Data Science undergrad at IIT Madras · AI/ML Research Intern at IIT Mandi
- Research, IIT Mandi — building a temporal action segmentation and classification pipeline on 9,000 human-manipulation videos. MediaPipe feature extraction, TCN/LSTM/MS-TCN models, for both real-time and offline action understanding. (Work is under double-blind review — repo isn't public yet, happy to discuss the approach directly.)
- RAG & agents — designing retrieval and tool-calling systems. See Movie Maven below.
- Computer vision — motion segmentation and trajectory analysis from video.
| Project | What it does | Stack |
|---|---|---|
| Movie Maven | RAG chatbot over a 45K+ movie dataset — hybrid retrieval combining metadata filtering, semantic search, and tool calling | LangGraph, Gemini, FAISS, Docker |
| Hand Trajectory Analysis & Motion Segmentation | Tracks 21 hand landmarks/frame, segments pick-and-place motion into phases via velocity-based valley detection — 71% baseline accuracy | MediaPipe, OpenCV, SciPy |
| Movie Recommendation System | Two-tower neural recommender learning user/movie embeddings from genre, year, and ratings | TensorFlow, Python |
Python PyTorch TensorFlow LangGraph FAISS OpenCV MediaPipe Flask PostgreSQL Docker
Open to internships in applied GenAI/agents or computer vision — reach out if something above is relevant to what you're building.