I build AI systems end to end — LLM-powered applications, agent tooling, and deep learning models that run behind real APIs instead of staying in notebooks.
Recent work includes a Gemini-powered resume analyzer served through FastAPI, an MCP server that exposes GitHub as structured tools to AI assistants, and a six-architecture benchmark for real-time sign language recognition.
- 🔭 Currently building LLM applications and agent tooling on top of the Model Context Protocol
- 🌱 Deepening my work in deep learning, model evaluation, and getting models into production
- 🎯 Open to AI / ML Engineering internships
- 📫 Reach me at rupakkumar76319@gmail.com
| Project | What it does | Stack |
|---|---|---|
| ATS Resume Analyzer | FastAPI service that parses PDF/DOCX resumes and scores them against a target job description using Gemini 2.5 Flash. Containerized with Docker. | FastAPI · Gemini API · Docker |
| GitHub Intelligence MCP Server | MCP server exposing GitHub as structured tools for AI assistants — repository analysis, issue management, and pull request insights. | Python · FastMCP · httpx |
| Sign Language Recognition | Real-time ASL recognition with MediaPipe Hands. Benchmarked 6 architectures (LSTM, GRU, BiLSTM, CNN, CNN-GRU, CNN-BiLSTM) across 36 classes — 92.68% best accuracy, 0.11s fastest inference. | TensorFlow · Keras · MediaPipe |
| Weathercast ML System | End-to-end next-day rainfall prediction — preprocessing, training, evaluation, and a served web interface for live predictions. | scikit-learn · XGBoost · Flask |
Also on my profile: an algorithmic trading backtester (moving-average crossover, Sharpe ratio, drawdown) and an e-commerce sales analysis over ~500K UK retail transactions.
Core — what I build with
Familiar — coursework and side work
- B.Tech, Electronics & Communication Engineering — Assam University, Silchar (ongoing)
- BS, Data Science and Applications — Indian Institute of Technology Madras (ongoing)