Freelance Python Developer & ML Engineer
July 2022 – Present • Remote- Developed predictive models and performed descriptive analytics for diverse datasets, generating actionable insights for clients.
- Built MapReduce NLP pipelines on Hadoop for high-efficiency word frequency analysis and large-scale text processing.
- Integrated large language models (LLMs) into production systems in collaboration with multi-disciplinary teams.
- Custom Object Detection and Tracking: Trained custom YOLO models for multi-class vehicle detection. Built a multi-lane vehicle tracking and counting system for real-time traffic analytics.
- Reinforcement Learning and Retrieval-Augmented Generation (RAG): Developed a complete RAG pipeline including RLHF, SFT, reward modeling, and deployment on Hugging Face. Implemented cloud-based RAG on Paperspace and AWS EC2 for scalable inference.
- Optimization of Object Sorting, Tracking, and Paddle Game Analysis: Improved multiple tracking pipelines using DeepSORT, FairMOT, and ByteTrack. Engineered an AI system for live paddle game analysis (ball tracking, shot detection, re-ID, gender detection). Trained LA Transformer re-ID models and YOLO models for multiple detection tasks.
- Created modular Dockerized services for open-source and closed-source components. Built internal APIs for video recording, dataset generation, and camera validation.
- AI Storyboard Video Generation Pipeline: Built T2I, I2I, and 3D asset pipelines using ComfyUI, Tripo SR, Gaussian Splatting, Hunyuan 3D, and Plan2Scene. Developed API endpoints for all workflows and implemented Blender bpy scripts for .glb scene generation.
- Agent ERP System: Deployed and configured multiple services via Docker (Postal, Chatwoot, Chartbrew, OsTicket, Documenso). Integrated SMTP, S3-compatible storage (Garage, Minio), and API proxies. Configured Higgs Audio (TTS/STT) and Voxtral for agent automation on AWS EC2. Set up Qwen Coder for code generation features.
- bCompetitive (Scraping, Automation, AI Integration) bcompetitive.ai – Designed and rebuilt browser automation strategies for Wayfair, Build.com, and Amazon scraping. Fixed the Recommended Price API pipeline by correcting malformed Celery → API request formatting. Currently building a Product Fetch API with improved normalization, caching, and scalable endpoints.