Optimize Deep Learning Projects on Upwork

I want to build an email classifier with two classes: approved and not approved. The labelled dataset is ready, and I need a NLP model to classify my emails. The project involves training a model for binary classification using the provided dataset. The ideal candidate should have experience in NLP and machine learning.

We are seeking an AI & ML Engineer to develop a predictive model for our SaaS platform. The role involves designing, training, and deploying models using Python, with a focus on improving customer retention. The ideal candidate will have experience in machine learning and data analysis, and be able to work independently. This is a Full-time engagement for 3 to 6 months.

We are looking for an AI Engineer to review and improve a small Python based RAG chatbot.

The chatbot currently uses OpenAI API, LangChain, embeddings, and a vector database to answer questions from uploaded documents. We need someone to identify the issue, improve document retrieval, and make the chatbot responses more accurate.

The selected developer should have experience with:

• Python and FastAPI
• OpenAI API and Large Language Models
• LangChain or LangGraph
• RAG pipeline development
• Embeddings and vector databases
• Pinecone, ChromaDB, FAISS, or pgvector
• Prompt engineering and response accuracy
• AI chatbot debugging and API integration

This is a small fixed price task for an experienced AI or Machine Learning Engineer. The initial budget is $25. More AI agent, RAG, Voice AI, and automation work may be available after successful completion.

Please briefly describe your experience with RAG chatbots and mention which vector databases you have worked with.

We are hiring an AI Engineer for a remote opportunity with our Airlines project. The ideal candidate should have hands-on experience building GenAI solutions, including RAG pipelines, vector embeddings, prompt engineering, MCP server development, and integrating multiple LLM providers. Experience working with AWS Neptune (Graph DB), OpenSearch (Vector Store), Redis, REST APIs, and SSE-based streaming services is required. Exposure to LangChain, MCPSharp, or ModelContextProtocol.SDK is a plus. If interested, please share your updated resume along with your total years of experience, years of GenAI experience, RAG experience, MCP/Agentic AI experience, current location, work authorization, and availability to start.

• How many years of experience do you have with RAG (Retrieval-Augmented Generation) development?
• How many years of experience do you have building AI orchestration / agentic workflows / MCP servers?

Agentic AI / RAG Engineer for Employment Law Assistant POC

Project Overview

We are looking for an experienced **AI/LLM Engineer** to build Version 1 of an AI-powered legal assistant focused on employment law.

The initial objective is to develop a functional proof of concept using **Agentic AI, Retrieval-Augmented Generation, and a legal knowledge database**. The platform should help users understand employment law, work with legal documents, receive dispute-resolution guidance, and access relevant legal updates in clear, understandable language.

This is not intended to be a generic chatbot. We need a developer who understands how to build a reliable, source-grounded AI system for a legal use case, with strong attention to accuracy, traceability, data privacy, and hallucination reduction.

## Core POC Features

The prototype should include:

* A conversational legal assistant focused on employment law
* Retrieval-Augmented Generation over legal documents, regulations, policies, precedents, and internal knowledge sources
* Agentic workflows for handling multi-step legal queries
* Clear explanations of complex legal terminology in plain language
* Source citations and references for generated answers
* Legal document upload, parsing, summarization, and question answering
* Context-aware follow-up questions when user input is incomplete
* Basic dispute-resolution and next-step guidance
* User interaction logging and feedback collection
* An evaluation mechanism for measuring answer relevance, faithfulness, and accuracy
* A secure admin workflow for adding, updating, and managing legal knowledge

## AI and Machine Learning Requirements

The AI system should:

1. Understand legal terminology and translate complex legal language into user-friendly explanations.

2. Use modern NLP and large language models to interpret user questions and generate contextually relevant responses.

3. Retrieve information from an approved legal database before producing legal answers.

4. Provide citations or references showing which legal sources were used.

5. Support multi-step reasoning and tool usage through an agentic architecture.

6. Improve over time through a controlled feedback and evaluation process.

We do not expect the model to retrain automatically on every user conversation. Instead, the preferred approach is to capture feedback and interactions, evaluate response quality, improve prompts and retrieval, update the knowledge base, and optionally create curated datasets for future fine-tuning.

## Preferred Technical Approach

We are open to architectural recommendations, but the preferred modern stack may include:

* Python
* FastAPI
* OpenAI, Anthropic Claude, Gemini, or suitable open-source LLMs
* LangGraph, LangChain, LlamaIndex, or a custom agent orchestration framework
* PostgreSQL with pgvector, Pinecone, Weaviate, Qdrant, or another vector database
* Hybrid semantic and keyword search
* Reranking models for improved retrieval accuracy
* Structured output and tool-calling workflows
* OCR and document processing for PDF, DOCX, and scanned legal files
* React or Next.js for the prototype interface
* AWS, Azure, or Google Cloud
* Docker and CI/CD
* LangSmith, Arize Phoenix, RAGAS, DeepEval, or similar LLM observability and evaluation tools

## Key Responsibilities

* Review the business concept and recommend the most suitable POC architecture
* Design the Agentic AI and RAG workflow
* Build ingestion pipelines for legal documents and structured legal data
* Implement chunking, metadata extraction, embeddings, hybrid retrieval, and reranking
* Develop a source-grounded conversational assistant
* Implement document analysis and question-answering workflows
* Add citations, confidence indicators, and fallback behavior
* Create guardrails for unsupported, ambiguous, or high-risk queries
* Build a basic web interface or integrate the AI backend with an existing platform
* Implement user feedback and response evaluation mechanisms
* Document the architecture, deployment process, and recommended roadmap for Version 2

## Important Considerations

Experience with the following will be highly valued:

* Legal AI, LegalTech, RegTech, HRTech, or compliance platforms
* Employment law knowledge systems
* RAG accuracy and hallucination reduction
* Knowledge graphs, legal taxonomies, and metadata-based retrieval
* Multi-agent or agentic workflow architecture
* LLM security, prompt injection protection, and access control
* Personally identifiable information and confidential document handling
* Human-in-the-loop review workflows
* LLM evaluation and observability
* Designing AI systems that provide information and guidance without making unsupported legal conclusions

## Expected Deliverables

* Working Version 1 POC
* Agentic AI and RAG backend
* Legal knowledge ingestion pipeline
* Searchable vector or hybrid knowledge database
* Conversational legal assistant
* Document upload and analysis workflow
* Source citations in responses
* Basic user interface or API integration
* Feedback and evaluation workflow
* Deployment to a cloud environment
* Technical documentation
* Recommendations, timeline, and cost estimate for production development

## Ideal Candidate

You should have hands-on experience building production or prototype systems involving:

* LLM applications
* Agentic AI
* RAG pipelines
* Vector databases
* Document processing
* Prompt engineering
* Tool calling and structured outputs
* Backend API development
* Cloud deployment
* AI evaluation and monitoring

Please include the following in your application:

1. Examples of relevant Agentic AI, RAG, LegalTech, or document intelligence projects
2. The technology stack you recommend for this POC
3. How you would reduce hallucinations and ensure answers are grounded in approved legal sources
4. How you would safely use user feedback to improve the system over time
5. Your estimated timeline and team composition
6. Any major technical or regulatory risks you would address during the POC

Please begin your proposal with the phrase **“Legal AI POC”** so we know you have reviewed the full description.

I have started to build out my agentic ai infrastructure using Claude.

Perhaps I must switch to moonshot ai.

I am Buidling a master agentic ai infrastructure that will serve as the backbone and agentic infrastructure (70%) of the operating system I would like to build.

The remaining 30% will be built on top of the agentic infrastructure as individual business verticals each with its own criteria and agentic ai design that would call the master agentic infrastructure as required.

The objective is to build 10 business verticals where I have a set of specific criteria for.

I have already started the process as mentioned on claude, set up my GitHub and supabase.

I need someone to pick up the infrastructure development and promoting and deliver the infrastructure, because I do not have time to work on it every day as in parallel I will be focusing on the business verticals.

I need you in detail:

How would you approach it differently?
How long it would take you to do?

I am building this as a start up offering based in Dubai.

Specifically the Dubai AI campus.

I need you to be very specific with your thoughts on how you will build the master agentic infrastructure with governance and compliance and data sovereignty accounted for.

I need you to detail the number of proposed agents you will build and highlight what their roles will be.

There are already many open source GitHub projects available that can be used quickly such as Rufflo for infrastructure.

The main verticals I am building for with the criteria include:

E-commerce
Compliance
Corporate knowledge hubs
Fashion
Fan engagement

Etc..

I appreciate if you feel the budget should be higher, please do pitch your desired budget, with a draft timeline and deliverables you will achieve.

Ideally I am looking for individual talent, not companies.

As mentioned this will be developed into a start up and I will be applying for funding in Dubai.

I am looking to scale with someone surly and would look for someone to join the journey early on too.

I look forward to your input.

Thank you

PROJECT CONTEXT

We are building a Computer Vision system for video analysis. We need to hire 2 profiles:
1. ARCHITECT (technical design, decisions, oversight)
2. DEVELOPER (implementation, coding, optimization)

This is a 6-8 week MVP project to build the FOUNDATION of our system (foundation + basic detection/tracking).

After Phase 1 validation, we may extend with future phases.

CRITICAL POINT

This project CANNOT be solved with:
- LLM (ChatGPT, Claude, Gemini)
- Generic vision APIs (Google Cloud, Azure)
- Low-code solutions or templates
- "Quick solutions"

This project REQUIRES:
- Real expertise in classical Computer Vision
- Deep Learning (PyTorch or TensorFlow)
- Proven experience: object detection + video tracking
- GPU optimization (NVIDIA CUDA)
- Production-grade code

TECHNICAL ARCHITECTURE

MVP Phase 1 = 2 main components:

1. VIDEO PIPELINE FOUNDATION
• Optimized video-to-frames decomposition
• Robust extraction and preprocessing
• Memory management for video streams
• Stack: OpenCV, FFmpeg, NumPy

2. DETECTION + TRACKING INTEGRATION
• YOLO v8+ for object/person detection
• DeepSORT for temporal tracking
• Frame-to-frame ID association
• Real-world condition reliability
• Stack: YOLO (Ultralytics), DeepSORT, PyTorch

MVP PHASE 1 TIMELINE

DURATION: 6-8 weeks total

WEEK 1-2: Architecture & Setup
- System design and technical decisions
- Development environment setup
- Testing framework design
- Performance benchmarking plan

WEEK 3-4: Video Pipeline
- Frame extraction pipeline
- Preprocessing and optimization
- Performance tests

WEEK 5-6: Detection + Tracking
- YOLO integration
- DeepSORT implementation
- ID management and stability

WEEK 7-8: Integration & Polish
- End-to-end testing
- Performance optimization
- Documentation and handoff
- Code cleanup and tests

DELIVERABLE: Working MVP with clean code + documentation

TECHNICAL STACK

Languages:
- Python 3.9+ (mandatory)
- C++ (optional, optimization)

Deep Learning & CV:
- PyTorch (preferred) or TensorFlow
- Ultralytics YOLO
- DeepSORT
- OpenCV, NumPy, scikit-image

Infrastructure:
- NVIDIA CUDA + cuDNN
- GPU memory optimization
- Docker (containerization)
- Git (version control)

WE'RE LOOKING FOR 2 PROFILES

PROFILE 1: ARCHITECT
- 5+ years Computer Vision experience
- Experience designing CV systems from scratch
- Strong understanding of YOLO, tracking algorithms
- Can make technical trade-off decisions
- Proposes optimizations and alternatives
- Oversees code quality and architecture
- Part-time or full-time OK (oversight role)

PROFILE 2: DEVELOPER (Senior)
- 3+ years production Computer Vision experience
- Expert in YOLO, OpenCV, PyTorch
- Experience with DeepSORT or similar tracking
- Strong Python + GPU optimization skills
- Can write clean, tested, production code
- Follows architectural decisions
- Full-time commitment preferred for 6-8 weeks

APPLICATION INSTRUCTIONS

Specify which role you're applying for: ARCHITECT or DEVELOPER

1. PORTFOLIO (mandatory)
- 2-3 CV projects you've delivered
- GitHub repos OR description + results
- Technologies used, your exact role

2. EXPERIENCE
- Years in Computer Vision (total)
- YOLO / object detection (versions)
- Tracking experience (DeepSORT, centroid, etc.)
- PyTorch/TensorFlow experience
- GPU optimization experience
- Production projects (not just Kaggle)

3. TECHNICAL OPINION
- Feasibility of MVP in 6-8 weeks: realistic?
- Any concerns or risks?
- Your approach to Phase 1?

4. TEAM PREFERENCE
- Have you worked with an architect/developer pair before?
- Preference for sync or async communication?

EXPECTED DELIVERABLES

CODE:
- GitHub repo (private/shared)
- Clean, tested Python code
- Unit tests (pytest)
- Requirements.txt / poetry.lock
- Meaningful docstrings

DOCUMENTATION:
- Architecture README (with diagrams)
- Installation guide
- API / usage examples
- Performance benchmarks (FPS, memory)
- Known limitations

HANDOFF:
- Code review completed
- Setup verified on clean machine
- Performance validated

FAQ

Q: Can one person do both (architect + developer)?
A: Possible for strong senior, but better with 2 people. More progress, better oversight.

Q: What YOLO version?
A: Open. v8 good starting point, v10 or ByteTrack alternatives welcome if justified.

Q: After MVP, what happens?
A: We evaluate quality. Strong MVP = potential Phase 2-4 continuation.

Let's build something solid!

• Describe your recent experience with similar projects

Job Title: Custom GPT Builder for Automotive Image Editing

I am looking for someone experienced with Custom GPTs and AI image editing to create a simple GPT for my automotive marketing business.

The GPT will need to:
● Accept an uploaded image of a vehicle.
● Remove the existing background.
● Place the vehicle onto a background image that I provide.
● Ensure the vehicle looks realistic and naturally positioned.
● Create realistic shadows beneath the vehicle.
● Match lighting and perspective between the vehicle and background.
● Remove sun glare, reflections and watermarks where possible.
● Clean up rough edges and improve overall image quality.

The goal is to produce professional, advertising-ready vehicle images for dealership websites, Carsales, Facebook and other digital advertising platforms.
I have attached three reference images:
1. A photo of a black Lexus taken in a dealership yard. The image contains shadows across the bonnet and windscreen which should be removed or significantly reduced.
2. A background image showing the garden setting that I would like used as the replacement background.
3. A reference image containing 24 before-and-after examples showing the style and quality of the finished result I am looking to achieve.

I will provide:
● Sample vehicle images
● Sample backgrounds
● Examples of the quality I am looking to achieve

This is a small trial project designed to test who can do a competent job, accurately follow my brief, and create the type of GPT I am looking for. I am specifically looking for someone to work with long-term. If I find the right person, there is potential for substantial ongoing work, including creating up to 50 similar GPTs in the future. At a minimum of $20 per GPT, this represents a considerable amount of additional work for the right candidate.

When applying, please: look
● Briefly explain your experience with Custom GPTs and AI image editing.
● Provide examples of similar work.
● Confirm your estimated time to complete the project.

As part of your application, please review the attached images and explain how you would approach creating a GPT that consistently produces results similar to the supplied examples.

We are seeking a skilled freelancer to enhance our existing (or newly created) machine learning model for detecting AI-generated text in Turkish. The ideal candidate will have experience in natural language processing and machine learning, with a focus on improving model accuracy and efficiency. The project involves working with Python and requires a strong understanding of AI-generated text detection techniques.

---
**What we have built:**
We have a working alpha system that detects AI-generated text in Turkish academic documents.

The current pipeline:
- Fine-tuned XLM-RoBERTa (xlm-roberta-base) on a labeled Turkish dataset
- Chunk-based inference with sliding window sentence scoring
- 94.2% accuracy on definitive decisions in production testing (Unfourtanetly real result is bad)
- Trained on 5,000 labeled text across 25 academic disciplines (human vs AI-generated)

**What we need:**

Improve the model's performance using one or more of the following approaches. You choose the best method based on results:

1. **Temperature scaling / calibration** — Make confidence scores meaningful and well-calibrated
2. **Perplexity-based signal (Binoculars approach)** — Add a training-free generator-agnostic signal using two open Turkish-capable LLMs (e.g. Qwen2.5). Fuse with existing classifier.
3. **Paraphrase augmentation (RADAR approach)** — Augment AI training samples with paraphrased versions to improve robustness against humanization tools
4. **Stronger backbone** — Evaluate mDeBERTa-v3-base or XLM-R-large as drop-in replacement
5. **Active learning** — Identify and prioritize the most informative uncertain examples for labeling

You may also build a brand-new model from scratch using the data. We accept that as well.

**Deliverables:**
- Python scripts only (no UI, no API, no deployment)
- Training script with the improved method
- Evaluation report: accuracy, FPR per category, comparison with baseline
- Saved model weights + inference script

**Data:**
- 5,000 labeled JSONL examples provided (title-matched human/AI pairs, 53 disciplines, 5 AI models)
- Can be increased if needed (by me if needed X10)
- Format: `"discipline", "title", "text" (human), "ai_text" (AI), "model"`

The most important point is this: we are testing with data that is not in the test set (i.e., not included in training). For example, we are testing with works written before 2020. Even though the scores appear high in Python's own internal test, the scores come out lower in real-world tests with data that was never part of the dataset. For that reason, "being done" does not mean the test performed in Python with the selected/curated data — it means a real-world test that the developer themselves can also conduct.

I have a dataset of natural-scene photographs and I need a reliable image classification pipeline that flags very specific visual features on each shot. The challenge is to teach a model to pick out those subtleties rather than just labeling broad object classes, so I’m looking for someone confident with convolutional networks, transfer learning, and the usual Python stack—PyTorch or TensorFlow with supporting libraries such as OpenCV, scikit-image, NumPy, and pandas for preprocessing and analysis.

You will receive the curated training and validation folders, along with a brief that explains exactly which visual cues matter to me. I’d like you to design, train, and fine-tune the network, then benchmark it with clear metrics (accuracy, precision/recall, confusion matrix). If data augmentation or additional sampling techniques are needed to improve generalization, feel free to implement them and document your rationale.

Deliverables:
• Clean, well-commented source code
• Trained model weights and an inference script that accepts a folder of images and outputs predictions
• A concise report summarizing architecture choices, training parameters, and evaluation results
• Quick-start instructions so I can reproduce the run on my own machine (GPU or CPU)

I value readable, maintainable code and straightforward explanations over flashy notebooks, so keep it pragmatic. Once we agree on acceptance criteria, I’m happy to move fast and iterate together until the classifier zeroes in on those features with solid confidence.

Skills: Python, Matlab and Mathematica, Algorithm, Machine Learning (ML), OpenCV, NumPy, Deep Learning, Convolutional Neural Network, YOLO

I’m building a small proof-of-concept that listens to an audio file and tells me how it feels. The focus is purely on analysis, not generation or recommendation. What I need is a working mood-detection pipeline that can ingest common formats (MP3, WAV, FLAC), extract the relevant features, and return an accurate emotional label for each track.

I’m open to either classical audio-signal processing combined with machine-learning classifiers, or a deep-learning approach built on libraries such as Librosa, PyTorch or TensorFlow—whichever you feel delivers the best accuracy without excessive latency. Please train or fine-tune your model on a dataset large enough to generalise across genres, then expose the result through a simple Python function or lightweight API endpoint.

Deliverable: fully commented source code, a short README explaining setup and usage, and a brief report on model performance (accuracy, key metrics, and any limitations you observed).

Skills: C Programming, Python, Software Architecture, Machine Learning (ML), C++ Programming, Audio Processing, Deep Learning, Classification

Location: United States
Description
Join a leading AI lab's cutting-edge GenAI team to be at the core of the AI revolution, where your expertise fuels the development of the most advanced Large Language Models.

Skills: Python, C++ Programming, Rust, Deep Learning, AI Research, AI Development

I need a concise yet clear live demonstration that introduces reinforcement learning through a Q&A-driven format. During the session you will walk through a working example—feel free to use Python, OpenAI Gym, or any comparable toolkit—and answer a short set of practical-application questions I’ll provide in advance.

Alongside the demo, complete a small written assignment that summarises the example, explains the key decisions you made, and responds to those same practical-application questions in text form (about one page is enough).

Deliverables
• Live demo link or recording (10-15 min)
• Source code/notebook used in the demo
• Written assignment (≈1 page) covering the Q&A

Keep the material beginner-friendly, focused on showing how reinforcement learning works in practice rather than delving into every algorithmic detail.

Skills: JavaScript, Python, Software Architecture, Machine Learning (ML), C++ Programming, Artificial Intelligence, Deep Learning, Reinforcement Learning

We're a professional trader building a private, AI-driven market research engine, and we're looking for a senior Python / machine learning engineer to act as Fractional CTO across a two-phase build.
Phase 1 (this contract): a Financial Market Signal Backtesting Dashboard — a clean measurement harness for trend, momentum, and multi-signal models with walk-forward backtesting.
Scope:
• Python project + GitHub repo
• Historical market-data ingestion (ETFs via yfinance / market-data APIs, SQL storage)
• Baseline models: 200-day trend, time-series & cross-sectional momentum, equal-weight multi-signal
• Walk-forward backtesting with look-ahead-bias controls
• Evaluation metrics: hit rate, Brier score, Sharpe, Sortino, Calmar, max drawdown, turnover, calibration
• A dashboard comparing signals and model performance
Stack: Python, pandas, NumPy, SQL, Python-based dashboard.
Phase 2 (follow-on): we extend the harness into the AI layer — an OODA / weight-of-evidence engine that scores competing market hypotheses. This involves LLM-based signal processing, an AI agent workflow over financial research and macro data, model evaluation / observability (LLMOps), and potentially fine-tuning and RAG over our research library. Strength here shapes who we keep on as Fractional CTO.
You are:
• A senior Python engineer with quantitative finance / backtesting experience
• Strong in machine learning and statistical evaluation
• Comfortable owning architecture as a Fractional CTO — evaluations, regression tests, observability, clean handover
• Bonus for Phase 2: AI agents, LLMs, LLMOps (LangChain / LangGraph, fine-tuning, RAG)

Project Description

We are building an early-stage AI-powered vehicle damage assessment startup.

We are looking for an experienced Computer Vision Engineer to develop the core AI engine for our prototype.

We only need:

* AI model training
* Cloud deployment
* API development

Project Scope

1. Prepare and clean vehicle damage datasets.
2. Fine-tune a YOLOv8 / YOLOv11 model (or equivalent) for vehicle damage detection.
3. Detect damaged vehicle parts and classify damage categories.
4. Generate structured JSON output.
5. Deploy the trained model to the cloud.
6. Build a FastAPI endpoint for inference.
7. Provide documentation for API integration.

Deliverables

* Trained model weights
* Source code
* Dataset preparation pipeline
* FastAPI endpoint
* Cloud deployment
* API documentation
* Basic testing/demo interface (optional)

Required Skills

* Python
* PyTorch
* YOLOv8 / YOLOv11
* Computer Vision
* Image Segmentation
* FastAPI
* Cloud Deployment

Nice to Have

* Automotive damage assessment experience
* Custom dataset training experience


How to Apply

Please start your proposal with the word:

“Prototype”

and include:

1. Similar computer vision projects.
2. Experience training custom YOLO models.
3. GitHub profile or portfolio.
4. Estimated timeline.
5. Your estimated budget range for this scope
6. Brief explanation of how you would approach the dataset and model training process.
We are looking for a long-term technical partner for future phases.

I have a curated dataset of patients’ genetic profiles and need a deep-learning solution that can reliably flag the presence of heart disease. Because the data are entirely genomic, the job begins with thoughtful preprocessing and feature engineering (handling high-dimensional SNPs, normalisation, train/validation split, class-imbalance techniques if required).

My single, overriding success metric is F1 Score; accuracy alone will not do, so the model must be tuned to balance precision and recall. You may choose the exact framework—PyTorch or TensorFlow/Keras are both fine—as long as the final code is clean, reproducible, and GPU-ready.

Deliverables
• End-to-end Python code or notebook that loads the raw genetic data, performs preprocessing, trains the deep neural network, and outputs predictions.
• Saved, versioned model weights plus a short README on how to retrain or fine-tune.
• Evaluation report that highlights F1 Score, precision, recall, and confusion matrix on the held-out test set.
• Brief architecture rationale so I understand why your chosen layers, regularisation, and hyperparameters suit genomic inputs.

I will provide the dataset and any relevant phenotype labels as soon as we start. My priority is a clear, well-documented workflow that I can deploy or extend later, so please keep modularity and readability in mind throughout the project.

Skills: Python, Matlab and Mathematica, Machine Learning (ML), Data Mining, Data Science, Neural Networks, Keras, Data Analysis, Deep Learning, Model Deployment

We need a developer to integrate AI features into our Shopify store. The system should recommend products based on customer behavior and improve conversions. Experience with AI APIs, Shopify custom apps, and automation workflows is required. The goal is to create a smarter shopping experience and increase sales.

WHAT WE'RE SEEKING
We are seeking a skilled developer to build an offline, real-time assessment system that analyses a user through webcam and microphone input.
The system should assess visible communication signals such as posture, gaze/head position, facial cues, hand movement, body movement, and voice delivery cues such as pitch, pace, pauses, volume, and intonation.
This is not a cartoon avatar, game character, VR experience, or VFX project. The core deliverable is the local assessment engine. The realistic human-facing interface is a separate presentation layer and should be implemented using the simplest practical method that meets the agreed requirements.

THE MISSION
Build a local computer vision and voice analysis system that processes live webcam and microphone input in real time, converts those signals into assessment-ready features, applies configurable assessment logic, and produces real-time feedback through a realistic human-facing interface.
The human-facing interface may use pre-recorded human video states, a lightweight talking-head interface, a voice-led coach interface, visual overlays, or another local method proposed by the developer.
All execution must run locally. No cloud inference, telemetry, analytics, external callouts, or subscription runtime dependencies are permitted.

TARGET DEVICE
Primary target: Windows 11 consumer hardware.
Must be benchmarked on the agreed target machine.
GPU acceleration may be used where appropriate.
Must not require non-consumer cluster GPUs.
Kinect, depth cameras, or specialist hardware may be proposed only as optional enhancements, not default requirements.

IN SCOPE

- Real-time webcam capture.
- Real-time microphone capture.
- Webcam-based pose estimation / body landmark tracking.
- Face landmark detection.
- Hand landmark tracking where practical.
- Gaze or head-pose estimation where practical.
- Voice feature extraction, including pitch, pace, pauses, volume, energy, and intonation features.
- Feature extraction layer that converts raw landmarks/audio into assessment signals.
- Configurable assessment logic using local JSON/YAML-style configuration.
- Hot reload for assessment logic without restarting the application.
- Real-time feedback events based on the assessment logic.
- Realistic human-facing interface layer.
- Visual debug overlay showing extracted landmarks/features during development and demo.
- Local benchmarking of latency, FPS, CPU/GPU usage, RAM usage, and sustained runtime.
- Offline verification showing no external calls.
- Clear separation between capture, feature extraction, assessment logic, scoring, feedback events, configuration, human interface, and application/UI layers.
- Documented interfaces between major components.
- Setup documentation, run commands, architecture notes, benchmark notes, and known limitations.

TECH STACK
Note: We will mutually agree on any tech stack changes to ensure no risk to target expectations and success factors:
* Visual engine: Unreal Engine 5
* Human character: MetaHuman first; custom rig later only if needed
* Face/lip animation: NVIDIA Audio2Face-3D / ACE Unreal Plugin
* Body/gaze/gesture: Unreal animation state machine
* Real-time audio/session: LiveKit Agents
* Turn-taking: LiveKit turn detector + Silero VAD
* TTS: Benchmark Chatterbox Turbo
* Backend: Existing backend via clean event bridge
* UI: React Tauri 1.8 wrapper - AJ to do vs PYQT6
* External webcam: OBS Virtual Camera only if needed
MediaPipe : Local multi-modal framework that extracts face, iris, hand, and body pose landmarks concurrently using device hardware.​
ONNX Runtime (onnxruntime-gpu): High-performance local inference engine optimized for running .onnx models utilizing Windows CUDA or DirectML acceleration.​
OpenCV (opencv-python): Native computer vision library used for real-time webcam frame acquisition, image preprocessing, and drawing visual debug overlays.​
sounddevice: Thread-safe audio library interfacing with Windows WASAPI/ASIO drivers to stream raw microphone input into NumPy arrays.​
Silero VAD: Ultra-lightweight voice activity detection model running locally via ONNX to segment real-time speech and calculate pauses.​
parselmouth (Praat for Python): Local acoustic analysis library utilizing mathematical signal processing to extract fundamental frequency for pitch tracking.​
scipy.signal & numpy: Scientific computing libraries used to perform instantaneous local RMS calculations for real-time volume and energy tracking.​
opensmile: Standardized feature extraction framework optimized in C++ to capture comprehensive acoustic parameter sets completely offline.​
PyQt6 / PySide6: C++ backed GUI framework providing hardware-accelerated widgets (QMediaPlayer) and thread-safe signaling to drive the presentation layer.​
watchdog: Local file monitoring utility that actively monitors file changes and triggers instantaneous hot-reloads of configuration files.​
psutil: System monitoring tool that queries the local OS kernel to benchmark CPU, RAM, and application process utilization.​
pynvml: Direct local wrapper for the NVIDIA Management Library used to programmatically log real-time GPU and VRAM metrics.

CLIENT TECHNICAL ENVIRONMENT:
* RTX 4080 15GB VRAM
* Ryzen 9 4950H
* 32GB RAM

PROCESSES (to be refined):
- User Experience During Session
app launch -- zoom like experience (see image)

- Post Session
User stats evaluation summary (driven by all of the info captured within this project): BPS, Pros, Cons, Body Language

Settings menu (icon) -
- Content : Load data (subject matter for user to discuss) - Rules config
- Presenter Persona : System Prompt
- LLM: Choose Ollama Model from drop-down list. User can download local models. Show local models downloaded. Can be deleted. Include synch button to fetch latest models/changes.



OUT OF SCOPE
- Cloud computer vision APIs.
- Cloud speech or voice analysis APIs.
- Subscription-based runtime dependencies.
- Telemetry, analytics, or external callouts.
- Enterprise production-grade deployment.
- Full 3D character pipeline unless separately approved.
- Skeletal mesh / rigged character development unless separately approved. (no only head, shoulder, arms only)
- VFX-heavy digital human rendering unless separately approved.
- Kinect or depth-camera dependency as a default requirement. (going into the distance)
- Deepfake-style synthesis unless explicitly approved in writing.
- Multilingual translation.
- Audio transcription beyond what is required for direct voice delivery analysis.
- Emotion recognition claims unless legally, technically, and contractually approved.
- Hardcoded assessment rules where configuration is practical.

IDEAL EXPERIENCE
Computer Vision
- Real-time pose estimation.
- Body landmark / skeletal landmark tracking.
- Face landmark detection.
- Hand tracking.
- Head-pose or gaze-related estimation.
- OpenCV, MediaPipe, ONNX Runtime, PyTorch, TensorRT, or equivalent local tooling.

Audio / Voice Analysis
- Real-time microphone handling.
- Pitch, tempo, pause, volume, energy, and intonation feature extraction.
- Local DSP/audio feature pipelines.
ffmpeg, PyAudio, sounddevice, librosa, torchaudio, or equivalent tooling.

Local ML / Runtime Optimisation
- Running models locally on consumer hardware.
- Optimising for latency and memory usage.
- Practical benchmarking and profiling.
- ONNX, PyTorch, TensorRT, or equivalent runtime choices.

Human Interface
- Video-based human presenter interfaces.
- Lightweight talking-head interfaces.
- Voice-led coaching interfaces.
- Real-time feedback UI.
- Ability to keep the presentation layer separate from the assessment engine.

Architecture
- Clean modular design.
- Testable component boundaries.
- Configuration-driven behaviour.
- Clear handover documentation.
- Practical trade-off decisions.
- Low technical debt MVP delivery.

NON-NEGOTIABLES
Offline Operation
- The system must run locally.
- No required internet connection.
- No telemetry.
- No analytics.
- No subscription runtime dependency.

Performance
- Target end-to-end feedback latency: under 100ms where technically practical.
- Candidate must benchmark actual latency on the agreed hardware.
- Candidate must benchmark memory usage and sustained runtime.
- Candidate must identify any parts that cannot realistically meet the target and explain why.

Architecture Quality
- The assessment engine and human-facing interface must be separate.
- Webcam capture, audio capture, feature extraction, assessment logic, configuration, feedback events, and presentation must not be tightly coupled.
- Assessment rules must be configurable where practical.
- The MVP must be maintainable, modular, and suitable for clean handover.

Communication
- Explain technical decisions clearly. - at least 30min video catch up, everyother day.
- Flag unclear requirements early.
- Flag delivery risks at least one day before they affect delivery.
- Shared-screen review sessions may be required.
- Live walkthroughs of architecture, benchmarks, and code may be required.
- Mutually agree any changes before committing

Output Quality
- AI-assisted coding is permitted up to a maximum of 60%.
- The developer remains responsible for all code quality.
- Code must be lean, modular, testable, and documented.
- No unsupported implementation choices.
- No unnecessary complexity.
- No hidden dependencies.

Handover
- Hand over all source code, assets, configuration files, setup notes, benchmark results, and documentation.
- Provide clear run instructions.
- Provide dependency and licensing notes.
- Provide known limitations.
- Remove project files and assets from supplier infrastructure after handover unless written approval is given.

DEFINITION OF DONE
The project is complete when:
- The application runs locally on the agreed Windows 11 hardware.
- Webcam and microphone input are captured and processed in real time.
- The system extracts pose/body landmarks, face landmarks, and agreed voice delivery features.
- The system extracts hand, gaze/head-pose, or equivalent visible communication features where technically practical.
- The system converts raw landmarks and audio features into assessment-ready signals.
- The system applies configurable local assessment logic.
Configuration changes can be hot reloaded without restarting the application.
- The system produces real-time feedback events. (BPS, body language, eye-tracking)
- The feedback events drive the agreed human-facing interface.
- The human-facing interface is separate from the assessment engine.
- The system includes a visual debug mode showing key extracted landmarks/features.
- Offline testing confirms no network calls (with exception of ollama cloud and initial system install), telemetry, analytics, or cloud dependencies.
- Source code is modular and documented.
- Handover package is complete.

ACCEPTANCE CRITERIA
- Live Demo
- Demonstrates webcam and microphone input being processed in real time.
- Shows extracted body/pose, face, and voice features.
- Shows real-time feedback being generated.
- Shows feedback driving the agreed human-facing interface.
- Configuration Demo
- Assessment logic is stored in local configuration files.
- A configuration value is changed during the demo.
- The system applies the changed logic without application restart.
- Offline Evidence
- Demo runs with network disabled.
- Logs or monitoring evidence show no external calls.
- No telemetry or analytics are present.
- Benchmark Evidence
- Benchmark report includes hardware used, model/runtime choices, FPS, latency, CPU/GPU usage, RAM usage, and sustained runtime.
- Any missed performance target is explained with evidence and mitigation options.
- Architecture Evidence
- Code review confirms separation between capture, feature extraction, assessment logic, configuration, feedback events, human interface, and application/UI layers.
- Interfaces between major components are documented.
- Architecture notes identify key trade-offs and known limitations.
- Source code, assets, configuration files, setup instructions, run commands, benchmark results, and known limitations are provided.
- Supplier confirms deletion of retained project materials unless written approval is given

PAYMENT CONDITIONS
* I’ll only accept work that meets the agreed standards. These will not change from what we've discussed

INVOICING
* Invoice like this please per week:
25hrs x 24.61 = 615GBP

Coding standards:
* Must have a clear and true seperation of concerns between FE and BE
* This means the front and backend are completely seperate and connected only by an integration layer.
* You can make superficial changes to the front layer, with zero risk to the backend.
* All files must be under 600 lines long inclusive of comments and single purpose. Only exception are shared utilities
* If a file exceeds 600 lines, then consider that it's either inefficient/bloated, or is doing more that one thing
* Examples of sinlge purpose: One file is for LLM integration, one file for feature x, another file for feature y (and so on)
* All code must include comments within files (applies only where possible, since you cannot add comments to some files due to syntax issues ). This includes:
* Before imports: file name with path + points on the unique features of the file
* Comments throughout code to explain what is dynamic such that any idiot like me can understand the code
* Ensure code is dynamic and flexible - not rigid, fragile or with hardcoded values .

I'm the founder of FretFlow, a guitar-learning app for beginners — a flowing, color-coded fretboard that shows you how to play the songs you love and slows down when you fall behind. The app is built and running on iPhone. I'm now building the feature at the heart of it, and I need a specialist for the one piece I can't build myself.

THE WORK
Build FretFlow's real-time audio recognition engine: it hears a guitar through the phone's microphone and identifies the chords and notes being played, live, on the device.

I want to start with a small, paid PROOF OF CONCEPT — a working demo that recognizes guitar chords (and ideally individual notes) in real time from mic input. If it works and we work well together, it grows into the full engine and a longer contract.

WHAT THE PROOF OF CONCEPT SHOULD SHOW
- Live chord recognition from a guitar through a microphone, in real time
- A clear path to running on-device on a phone (Core ML / TensorFlow Lite), low latency — not server-dependent

YOU'RE A FIT IF YOU'VE:
- Built audio/music ML focused on RECOGNITION — chord detection, pitch/note detection, music information retrieval (NOT music generation)
- Run audio ML models ON a mobile device (Core ML, TensorFlow Lite, or ONNX)
- Worked with real-time, low-latency audio

NICE TO HAVE: React Native / Expo familiarity; tools like Essentia, librosa, CREPE, or basic-pitch; you play guitar.

ABOUT THE STAGE
Early-stage, founder-led, pre-launch. This is a paid contract starting with the proof-of-concept, with real room to continue. I'm looking for someone excited to build the brain of a product from the ground up.

Please answer the screening questions below — proposals that show specific chord/pitch-recognition and on-device work go to the top of my list.

1. Have you built a chord or note/pitch recognition system before? Briefly describe it and share a link if you can.

2. Have you run an audio ML model on a mobile device (Core ML, TensorFlow Lite, or ONNX)? Which one, and roughly what latency did you get?

3. In a sentence or two, how would you approach real-time chord recognition from a phone's microphone?

Project specifications:
1. Cybersecurity and digital identily check, blockchain identity implementation.
2. Open APIs screaning and scraping. Very important!
3. AI/BCI/software related to my identity check, including any AI agent/MLM/LLM/digital twin..., background check.

We are looking for a remote Machine Learning Engineer to help build and deploy production document-AI systems. The work centres on optical character recognition (OCR), document analysis, computer vision, and NLP/LLM components that automate document-heavy operational processes. You will develop and optimise deep-learning models, build end-to-end ML pipelines (data ingestion, feature engineering, training, evaluation, inference), and deploy them using Docker, Kubernetes, and modern MLOps practices. Essential skills: strong Python; TensorFlow and/or PyTorch; OpenCV; and hands-on experience with OCR and computer vision. Familiarity with LLMs and document-analysis techniques (e.g., TrOCR) is a strong plus. This is an ongoing, fully remote contract role with flexible hours.

Designed and developed an AI-driven cardiac screening system that analyzes heart sound recordings to assist in early detection of cardiac abnormalities. Built a complete machine learning pipeline involving audio preprocessing, feature extraction, model training, and evaluation. Extracted MFCC-based acoustic features from heart sound signals and compared multiple classification algorithms including KNN, SVM, and Random Forest to optimize performance. Developed a web-based application, Heartify Pro, integrating machine learning predictions, automated clinical report generation, and a Llama-powered AI voice assistant for interactive healthcare guidance.

Technologies: Python, Scikit-learn, Librosa, MFCC, Flask, HTML/CSS/JavaScript, Llama API

Skills: Python, Artificial Intelligence, Deep Learning, Time Series Forecasting, Long Short-Term Memory Network

I want a compact AI solution that can watch a live camera feed, recognise people, vehicles and animals, then overlay on-screen the class label and an accurate distance reading for every object it sees. The system must work reliably indoors and outdoors without me having to switch profiles manually.

Here is what I expect:
• Real-time detection and tracking of the three classes mentioned above.
• A distance-estimation module (stereo, monocular depth, LiDAR fusion—whichever you judge most practical) that outputs metres or feet beside each bounding box.
• Clear text overlay on the video frame; no audible alerts are required.
• A single-command way to launch the application on a standard Windows or Linux laptop with an external or built-in camera.
• Source code (Python preferred), pre-trained weights, a short README for setup, and a calibration routine so I can adapt the model to new lenses.

Acceptance criteria
1. When an object is centred at 3 m, the displayed distance should be within ±10 %.
2. Frame rate should stay above 15 fps at 720p on a mid-range CPU/GPU.
3. Only the three specified classes appear; all else is ignored or labelled “other”.

Feel free to build on YOLOv8, OpenCV, PyTorch, depth-AI, or similar—just keep the install steps streamlined. Demo footage proving indoor and outdoor accuracy will help me verify results quickly.

Skills: C Programming, Python, C++ Programming, Arduino, OpenCV, Computer Vision, Deep Learning, YOLO

Project Title:
Development of a Specialized Text-to-Speech (TTS) Phonetic Dictionary and Lexicon for Singaporean English

Company Overview:
We are an innovative AI organization expanding our proprietary Voice AI capabilities. Our current system architecture is built on the Google Cloud Ecosystem (including Google AI Studio, Speech-to-Text, and Text-to-Speech Developer tools). We are now looking to hyper-localize our user experience by creating a highly accurate, custom pronunciation dictionary tailored specifically to the Singaporean market.

Project Scope & Objective:
We require a qualified Subject Matter Expert (SME) to design, build, and validate a comprehensive text-to-speech phonetic lexicon/dictionary for Singaporean English (SgE). This database will map localized vocabulary, loanwords, unique idioms, names, and structural expressions to their precise phonetic transcriptions—enabling our voice agent to sound authentically and naturally Singaporean.

The successful contractor will develop a master dictionary including:

Segmental Layer: Accurate phoneme mapping handling typical SgE vowel mergers (e.g., fleece/kit, dress/trap) and consonant structures.

Loanword Integration: Phonetic transcriptions of widespread substrate-derived loanwords (from Hokkien, Mandarin, Malay, and Tamil) commonly integrated into everyday Singaporean business and casual speech.

Suprasegmental/Prosodic Markers: Alignment with the syllable-timed rhythm and distinct intonational patterns characteristic of Singapore English.

Key Deliverables:
Phonetic Lexicon Database: A clean, machine-readable dataset (JSON, XML, or CSV format) containing words mapped to standardized phonetic alphabets accepted by modern TTS architectures (e.g., SAMPA, IPA, or Arpabet modified for SgE).

Grapheme-to-Phoneme (G2P) Ruleset: Documentation of structural linguistic rules to help the engine handle out-of-vocabulary (OOV) Singaporean terms seamlessly.

Integration Support: Collaborating briefly with our engineering team to ensure the lexical database uploads smoothly into our Google Cloud / Google AI Studio pipeline using custom voice tuning or Synthesis Markup Language (SSML) lexicons.

Required Qualifications & Certifications:
Applicants must possess formalized credentials in this domain. Please do not apply if you do not meet the following criteria:

Educational Certification: A Master’s Degree or Ph.D. in Linguistics, Phonetics, Computational Linguistics, or a closely related field.

Professional Expertise: Proven experience in lexicography, phonetic transcription, or building Lexicons specifically for speech synthesis (TTS) or automated speech recognition (ASR).

Localization Knowledge: Native or near-native familiarity with the linguistic nuances, phonology, and sociolinguistics of Singaporean English and Singlish.

Technical Familiarity: Understanding of speech technology frameworks and formatting data to be compatible with cloud speech architectures (experience with Google Cloud Speech tools is highly preferred).

• Please list any certifications related to this project
• Describe your recent experience with similar projects

Build an AI-Powered Demand Forecasting Application

Develop an end-to-end Demand Forecasting application capable of ingesting historical and external demand drivers, performing automated data preparation, generating forecasts using multiple statistical and AI/ML models, and comparing model performance to identify the most accurate forecasting approach.

1. Data Inputs
The application should support importing historical demand data from Excel (.xlsx) and CSV files.

A. Transactional and Supply Chain Data
Sales Orders from Order Management systems

Item Master Attributes

Customer Attributes

Organization/Business Unit Attributes

Inventory Levels

Stockout History

Supplier Lead Times

Production Capacity Constraints

Supplier Disruptions

Logistics and Transportation Delays

B. Event-Based Demand Drivers
Examples:

Christmas

Diwali

Thanksgiving

Black Friday

Cyber Monday

Super Bowl

School Openings

Pay Cycles

Month-End Demand Surges

Quarter-End Demand Surges

Seasonal Peaks

Fiscal Calendar Events

C. Weather and Environmental Factors
Examples:

Temperature (HVAC demand)

Rainfall (umbrella sales)

Humidity (beverage demand)

Storm Alerts (panic buying)

Air Quality Index (healthcare products)

Climate Anomalies

D. Economic and Market Indicators
Examples:

Inflation Rates

Consumer Spending

Interest Rates

GDP Growth

Fuel Prices

Currency Exchange Rates

Import/Export Trends

Consumer Confidence Index

Industry-Specific Market Indicators

E. Digital and Behavioral Signals
Examples:

Website Traffic

Search Trends

Clickstream Data

Product Page Views

Cart Additions

Social Media Sentiment

Product Ratings and Reviews

Online Demand Signals

2. Data Preparation and Cleansing
The application should automatically perform:

Missing Value Detection and Imputation

Outlier Detection and Correction

Regime Change Identification

Trend and Seasonality Analysis

Data Normalization and Scaling

Feature Engineering

Time-Series Decomposition

Demand Segmentation

Data should be classified and analyzed across:

Item Dimension

Customer Dimension

Organization Dimension

Geography Dimension

Product Hierarchy

3. Forecast Frequency Optimization
The system should automatically evaluate and recommend the most appropriate forecasting granularity:

Daily Forecasts

Weekly Forecasts

Monthly Forecasts

The recommendation should be based on historical forecast accuracy, demand volatility, seasonality patterns, and business requirements.

4. Forecasting Models
The application should train, evaluate, and compare multiple forecasting models, including:

Statistical Models
ARIMA

SARIMA

Machine Learning Models
XGBoost Regressor

Prophet

Deep Learning Models
LSTM

Attention-Based LSTM

Stacked LSTM

Bidirectional LSTM

Quantum-Inspired LSTM

The platform should:

Train models using configurable epochs

Automatically tune hyperparameters

Optimize model weights

Perform cross-validation

Select the best-performing model based on evaluation metrics

5. Model Evaluation Metrics
Generate a comprehensive comparison report using:

Metric
Accuracy (%)
Precision (%)
Recall (%)
F1 Score (%)
MAE (Mean Absolute Error)
RMSE (Root Mean Squared Error)
MAPE (Mean Absolute Percentage Error)
Bias
Forecast Value Added (FVA/FAV)
Output Example:

Model Accuracy (%) Precision (%) Recall (%) F1 Score (%) MAE RMSE MAPE Bias FVA
6. Forecast Generation and Explainability
For each forecast, provide:

Forecast Quantity

Confidence Intervals

Predicted Trend Direction

Key Demand Drivers

Feature Importance Rankings

Event Impact Analysis

Weather Impact Analysis

Economic Impact Analysis

7. Outputs
The application should export results to:

Excel (.xlsx)

CSV

Output files should include:

Forecast Results

Model Performance Comparison

Forecast Accuracy Metrics

Feature Importance Analysis

Demand Driver Contributions

Recommended Forecasting Frequency (Daily/Weekly/Monthly)

8. User Experience Requirements
Upload historical data through Excel or CSV files.

Configure forecast horizon (30, 60, 90, 180, 365 days).

Select demand drivers to include.

Automatically run all forecasting models.

Compare results side-by-side.

Recommend the best model based on performance metrics.

Generate downloadable forecast reports and visual dashboards.

Support future integration with ERP systems such as Oracle Fusion Cloud, SAP, and Dynamics 365.

Goal: Build an enterprise-grade AI Demand Forecasting platform that combines statistical forecasting, machine learning, deep learning, external demand sensing signals, and automated model selection to deliver highly accurate, explainable, and scalable demand forecasts.

Skills: Python, Artificial Intelligence, Deep Learning, Time Series Forecasting, Long Short-Term Memory Network

WHAT WE'RE SEEKING

We are seeking a skilled developer to build an offline, real-time assessment system that analyses a user through webcam and microphone input.

The system should assess visible communication signals such as posture, gaze/head position, facial cues, hand movement, body movement, and voice delivery cues such as pitch, pace, pauses, volume, and intonation.

This is not a cartoon avatar, game character, VR experience, or VFX project. The core deliverable is the local assessment engine. The realistic human-facing interface is a separate presentation layer and should be implemented using the simplest practical method that meets the agreed requirements.

THE MISSION

Build a local computer vision and voice analysis system that processes live webcam and microphone input in real time, converts those signals into assessment-ready features, applies configurable assessment logic, and produces real-time feedback through a realistic human-facing interface.

The human-facing interface may use pre-recorded human video states, a lightweight talking-head interface, a voice-led coach interface, visual overlays, or another local method proposed by the developer.

All execution must run locally. No cloud inference, telemetry, analytics, external callouts, or subscription runtime dependencies are permitted.

TARGET DEVICE

Primary target: Windows 11 consumer hardware.
Must be benchmarked on the agreed target machine.
GPU acceleration may be used where appropriate.
Must not require non-consumer cluster GPUs.
Kinect, depth cameras, or specialist hardware may be proposed only as optional enhancements, not default requirements.

IN SCOPE

- Real-time webcam capture.
- Real-time microphone capture.
- Webcam-based pose estimation / body landmark tracking.
- Face landmark detection.
- Hand landmark tracking where practical.
- Gaze or head-pose estimation where practical.
- Voice feature extraction, including pitch, pace, pauses, volume, energy, and intonation features.
- Feature extraction layer that converts raw landmarks/audio into assessment signals.
- Configurable assessment logic using local JSON/YAML-style configuration.
- Hot reload for assessment logic without restarting the application.
- Real-time feedback events based on the assessment logic.
- Realistic human-facing interface layer.
- Visual debug overlay showing extracted landmarks/features during development and demo.
- Local benchmarking of latency, FPS, CPU/GPU usage, RAM usage, and sustained runtime.
- Offline verification showing no external calls.
- Clear separation between capture, feature extraction, assessment logic, scoring, feedback events, configuration, human interface, and application/UI layers.
- Documented interfaces between major components.
- Setup documentation, run commands, architecture notes, benchmark notes, and known limitations.

OUT OF SCOPE

- Cloud computer vision APIs.
- Cloud speech or voice analysis APIs.
- Subscription-based runtime dependencies.
- Telemetry, analytics, or external callouts.
- Enterprise production-grade deployment.
- Full 3D character pipeline unless separately approved.
- Skeletal mesh / rigged character development unless separately approved.
- VFX-heavy digital human rendering unless separately approved.
- Kinect or depth-camera dependency as a default requirement.
- Deepfake-style synthesis unless explicitly approved in writing.
- Multilingual translation.
- Audio transcription beyond what is required for direct voice delivery analysis.
- Emotion recognition claims unless legally, technically, and contractually approved.
- Hardcoded assessment rules where configuration is practical.

IDEAL EXPERIENCE

Computer Vision

- Real-time pose estimation.
- Body landmark / skeletal landmark tracking.
- Face landmark detection.
- Hand tracking.
- Head-pose or gaze-related estimation.
- OpenCV, MediaPipe, ONNX Runtime, PyTorch, TensorRT, or equivalent local tooling.

Audio / Voice Analysis

- Real-time microphone handling.
- Pitch, tempo, pause, volume, energy, and intonation feature extraction.
- Local DSP/audio feature pipelines.
ffmpeg, PyAudio, sounddevice, librosa, torchaudio, or equivalent tooling.

Local ML / Runtime Optimisation

- Running models locally on consumer hardware.
- Optimising for latency and memory usage.
- Practical benchmarking and profiling.
- ONNX, PyTorch, TensorRT, or equivalent runtime choices.

Human Interface

- Video-based human presenter interfaces.
- Lightweight talking-head interfaces.
- Voice-led coaching interfaces.
- Real-time feedback UI.
- Ability to keep the presentation layer separate from the assessment engine.

Architecture

- Clean modular design.
- Testable component boundaries.
- Configuration-driven behaviour.
- Clear handover documentation.
- Practical trade-off decisions.
- Low technical debt MVP delivery.

NON-NEGOTIABLES

Offline Operation

- The system must run locally.
- No required internet connection.
- No cloud inference.
- No telemetry.
- No analytics.
- No external service calls.
- No subscription runtime dependency.

Performance

- Target end-to-end feedback latency: under 100ms where technically practical.
- Candidate must benchmark actual latency on the agreed hardware.
- Candidate must benchmark memory usage and sustained runtime.
- Candidate must identify any parts that cannot realistically meet the target and explain why.

Architecture Quality

- The assessment engine and human-facing interface must be separate.
- Webcam capture, audio capture, feature extraction, assessment logic, configuration, feedback events, and presentation must not be tightly coupled.
- Assessment rules must be configurable where practical.
- The MVP must be maintainable, modular, and suitable for clean handover.

Communication

- Explain technical decisions clearly.
- Flag unclear requirements early.
- Flag delivery risks at least one day before they affect delivery.
- Shared-screen review sessions may be required.
- Live walkthroughs of architecture, benchmarks, and code may be required.

Output Quality

- AI-assisted coding is permitted up to a maximum of 60%.
- The developer remains responsible for all code quality.
- Code must be lean, modular, testable, and documented.
- No unsupported implementation choices.
- No unnecessary complexity.
- No hidden dependencies.

Handover

- Hand over all source code, assets, configuration files, setup notes, benchmark results, and documentation.
- Provide clear run instructions.
- Provide dependency and licensing notes.
- Provide known limitations.
- Remove project files and assets from supplier infrastructure after handover unless written approval is given.

DEFINITION OF DONE

The project is complete when:

- The application runs locally on the agreed Windows 11 hardware.
- Webcam and microphone input are captured and processed in real time.
- The system extracts pose/body landmarks, face landmarks, and agreed voice delivery features.
- The system extracts hand, gaze/head-pose, or equivalent visible communication features where technically practical.
- The system converts raw landmarks and audio features into assessment-ready signals.
- The system applies configurable local assessment logic.
Configuration changes can be hot reloaded without restarting the application.
- The system produces real-time feedback events.
- The feedback events drive the agreed human-facing interface.
- The human-facing interface is separate from the assessment engine.
- The system includes a visual debug mode showing key extracted landmarks/features.
- Benchmark results show latency, FPS, CPU/GPU usage, RAM usage, and sustained runtime.
- Offline testing confirms no network calls, telemetry, analytics, or cloud dependencies.
- Source code is modular and documented.
- Handover package is complete.

ACCEPTANCE CRITERIA

- Live Demo
- Demonstrates webcam and microphone input being processed in real time.
- Shows extracted body/pose, face, and voice features.
- Shows real-time feedback being generated.
- Shows feedback driving the agreed human-facing interface.
- Configuration Demo
- Assessment logic is stored in local configuration files.
- A configuration value is changed during the demo.
- The system applies the changed logic without application restart.
- Offline Evidence
- Demo runs with network disabled.
- Logs or monitoring evidence show no external calls.
- No telemetry or analytics are present.
- Benchmark Evidence
- Benchmark report includes hardware used, model/runtime choices, FPS, latency, CPU/GPU usage, RAM usage, and sustained runtime.
- Any missed performance target is explained with evidence and mitigation options.
- Architecture Evidence
- Code review confirms separation between capture, feature extraction, assessment logic, configuration, feedback events, human interface, and application/UI layers.
- Interfaces between major components are documented.
- Architecture notes identify key trade-offs and known limitations.
- Handover Evidence
- Source code, assets, configuration files, setup instructions, run commands, dependency notes, licensing notes, benchmark results, and known limitations are provided.
- Supplier confirms deletion of retained project materials unless written approval is given.

• What similar real-time CV, pose estimation, body tracking, voice analysis, or human-interface work you have already completed
• Explain the simplest practical MVP approach you recommend.
• What tools/models/runtimes you propose and why.
• How you would prove offline operation. How you would benchmark latency and memory usage. What are the main technical risks.
• Whether you would use webcam-based pose estimation, depth-camera tracking, or another method.

Build an AI-Powered Demand Forecasting Application

Develop an end-to-end Demand Forecasting application capable of ingesting historical and external demand drivers, performing automated data preparation, generating forecasts using multiple statistical and AI/ML models, and comparing model performance to identify the most accurate forecasting approach.

1. Data Inputs
The application should support importing historical demand data from Excel (.xlsx) and CSV files.

A. Transactional and Supply Chain Data
Sales Orders from Order Management systems

Item Master Attributes

Customer Attributes

Organization/Business Unit Attributes

Inventory Levels

Stockout History

Supplier Lead Times

Production Capacity Constraints

Supplier Disruptions

Logistics and Transportation Delays

B. Event-Based Demand Drivers
Examples:

Christmas

Diwali

Thanksgiving

Black Friday

Cyber Monday

Super Bowl

School Openings

Pay Cycles

Month-End Demand Surges

Quarter-End Demand Surges

Seasonal Peaks

Fiscal Calendar Events

C. Weather and Environmental Factors
Examples:

Temperature (HVAC demand)

Rainfall (umbrella sales)

Humidity (beverage demand)

Storm Alerts (panic buying)

Air Quality Index (healthcare products)

Climate Anomalies

D. Economic and Market Indicators
Examples:

Inflation Rates

Consumer Spending

Interest Rates

GDP Growth

Fuel Prices

Currency Exchange Rates

Import/Export Trends

Consumer Confidence Index

Industry-Specific Market Indicators

E. Digital and Behavioral Signals
Examples:

Website Traffic

Search Trends

Clickstream Data

Product Page Views

Cart Additions

Social Media Sentiment

Product Ratings and Reviews

Online Demand Signals

2. Data Preparation and Cleansing
The application should automatically perform:

Missing Value Detection and Imputation

Outlier Detection and Correction

Regime Change Identification

Trend and Seasonality Analysis

Data Normalization and Scaling

Feature Engineering

Time-Series Decomposition

Demand Segmentation

Data should be classified and analyzed across:

Item Dimension

Customer Dimension

Organization Dimension

Geography Dimension

Product Hierarchy

3. Forecast Frequency Optimization
The system should automatically evaluate and recommend the most appropriate forecasting granularity:

Daily Forecasts

Weekly Forecasts

Monthly Forecasts

The recommendation should be based on historical forecast accuracy, demand volatility, seasonality patterns, and business requirements.

4. Forecasting Models
The application should train, evaluate, and compare multiple forecasting models, including:

Statistical Models
ARIMA

SARIMA

Machine Learning Models
XGBoost Regressor

Prophet

Deep Learning Models
LSTM

Attention-Based LSTM

Stacked LSTM

Bidirectional LSTM

Quantum-Inspired LSTM

The platform should:

Train models using configurable epochs

Automatically tune hyperparameters

Optimize model weights

Perform cross-validation

Select the best-performing model based on evaluation metrics

5. Model Evaluation Metrics
Generate a comprehensive comparison report using:

Metric
Accuracy (%)
Precision (%)
Recall (%)
F1 Score (%)
MAE (Mean Absolute Error)
RMSE (Root Mean Squared Error)
MAPE (Mean Absolute Percentage Error)
Bias
Forecast Value Added (FVA/FAV)
Output Example:

Model Accuracy (%) Precision (%) Recall (%) F1 Score (%) MAE RMSE MAPE Bias FVA
6. Forecast Generation and Explainability
For each forecast, provide:

Forecast Quantity

Confidence Intervals

Predicted Trend Direction

Key Demand Drivers

Feature Importance Rankings

Event Impact Analysis

Weather Impact Analysis

Economic Impact Analysis

7. Outputs
The application should export results to:

Excel (.xlsx)

CSV

Output files should include:

Forecast Results

Model Performance Comparison

Forecast Accuracy Metrics

Feature Importance Analysis

Demand Driver Contributions

Recommended Forecasting Frequency (Daily/Weekly/Monthly)

8. User Experience Requirements
Upload historical data through Excel or CSV files.

Configure forecast horizon (30, 60, 90, 180, 365 days).

Select demand drivers to include.

Automatically run all forecasting models.

Compare results side-by-side.

Recommend the best model based on performance metrics.

Generate downloadable forecast reports and visual dashboards.

Support future integration with ERP systems such as Oracle Fusion Cloud, SAP, and Dynamics 365.

Goal: Build an enterprise-grade AI Demand Forecasting platform that combines statistical forecasting, machine learning, deep learning, external demand sensing signals, and automated model selection to deliver highly accurate, explainable, and scalable demand forecasts.

Skills: Python, Artificial Intelligence, Deep Learning, Time Series Forecasting, Long Short-Term Memory Network

I have an open-source baseline in mind (the Ipernov face-swap repository) and need a truly real-time desktop application built from that starting point. The core of the job is to fork the repo, strip out any unnecessary overhead, and rebuild the inference pipeline so that live-streaming and video-call scenarios run smoothly with minimal latency and zero dropped frames.

What matters most
• Desktop focus – the executable must run locally on a typical gaming-class PC without cloud round-trips.
• Target Hardware Baseline – The application must be optimized to achieve these metrics natively on an RTX 4060 Laptop GPU (8GB VRAM) and an i7-14650HX CPU with 16GB RAM.
• Full face swap is the single non-negotiable feature; partial swaps or expression transfer are optional nice-to-haves if they do not compromise performance.
• Real-time speed – think webcam input at 720p/1080p and a sustained 30 FPS or better. This will involve CUDA, TensorRT/ONNX optimisations, lightweight encoder/decoder tweaks, and clever GPU memory management to ensure stable, long-term live streaming within a strict 8GB VRAM budget without memory leaks.
• Seamless output – the program should expose a virtual webcam so that OBS, Zoom, or Discord can pick it up instantly.

Deliverables
1. Clean forked codebase with all performance improvements clearly documented.
2. One-click desktop build (installer or packaged binaries).
3. Benchmark report showing FPS, latency, and system specs used.
4. Quick-start README plus comments explaining every critical optimisation step.

Testing & Sign-off
I will test the build on my own machine and sign off once the metrics match the benchmark you provide. If you enjoy squeezing out every last millisecond of inference time, this will be a fun project.

To Apply:
Please start your proposal with the words "Low Latency Achieved" so I know you have read this entire specification. In your proposal, briefly highlight your specific experience optimizing ONNX Runtime, TensorRT, or custom C++/Python pipelines for real-time video processing.

Skills: Python, CUDA, Machine Learning (ML), Computer Vision, Deep Learning, AI Rendering, AI Text-to-video, AI Chatbot Development, AI Model Development, AI Development

I have a sizeable EEG dataset that needs to be transformed into a working epileptic-seizure detection and short-term prediction pipeline. The workflow I have in mind is clear:

• Data preprocessing: begin with classic EEG Filtering and follow with signal Normalization to standardise every trial.

• Feature extraction: apply a Wavelet Transform to capture the relevant time–frequency patterns that precede a seizure.

• Dimensionality reduction: reduce the resulting feature space intelligently so the model is both fast and interpretable.

• Modelling: my first choice is a Random Forest classifier/regressor for both detection and prediction tasks; however, if you can show convincing evidence that a complementary deep-learning block will boost performance without sacrificing interpretability, I am open to discussion.

• Explainable AI: integrate SHAP or LIME so each prediction can be traced back to the contributing EEG channels, scales, or time-windows.

• Performance evaluation: deliver detailed metrics (accuracy, sensitivity, specificity, F1, ROC-AUC, false-alarm rate per hour) on a solid cross-validation split and, if possible, an unseen hold-out set.

Deliverables must include
– Clean, well-commented source code (Python preferred: NumPy, Pandas, MNE, scikit-learn, SHAP/LIME, etc.)
– A concise technical report that documents preprocessing choices, feature engineering, modelling decisions, and performance tables/plots
– Reproducible experiments packaged in notebooks or scripts with clear run instructions
– A read-me explaining environment setup and how to deploy or further train the system.

I value transparent, modular code, reproducibility, and plain-language explanations of the model’s decisions. If this matches your expertise, let’s make EEG more predictable together.

Skills: Python, Matlab and Mathematica, Algorithm, Statistics, Machine Learning (ML), Statistical Analysis, Data Science, Artificial Intelligence

NovaHealth is a doctor–patient matching app/platform that curates a small, high‑quality set of options instead of a giant directory. Each user conveys the service they need and the “care vibe” they prefer through an MBTI style form; the system then surfaces 3 tailored doctors with short intro videos, availability, and reputation signals so choosing feels simple instead of overwhelming.

I am looking for someone to create this AI matching system and the web app overall.

Seeking: Former Scale AI, Mercor, Surge, Invisible, Outlier, Handshake or similar

We’re building infrastructure that enables talent networks, universities, recruiters, tutoring platforms, and expert communities to participate in the AI workforce economy.

We already have access to a large database of technical and specialist talent. We’re interested in building the systems that can identify the right people with the lowest possible friction, deploy them into AI projects, and create a scalable bridge between talent ecosystems and AI labs.

We’re looking for someone who understands:

* AI data operations
* Workforce deployment at scale
* Quality control and evaluation frameworks
* Vendor and client relationships
* AI lab buying processes
* Building AI-first talent infrastructure

You might have experience at Scale AI, Mercor, Surge, Invisible, Outlier, Handshake, or a similar organisation.

We’re particularly interested in learning where the market is heading, overlooked opportunities, and how best to build the infrastructure, partnerships, and ecosystem needed to support the next generation of AI work.

• If you joined as an advisor, what are the three organisations you would speak to first and why?
• If you had access to 30,000 technical candidates tomorrow, what would be your plan to land the first paying Al customer in the next 90 days?
• Describe your recent experience with similar projects