Computer Science Engineering undergrad at Thapar Institute of Engineering and Technology (2023–2027, CGPA 9.06). I build ML, NLP and GenAI systems, from training models in PyTorch to shipping retrieval-augmented LLM apps with FastAPI.
ResearchHelp · RAG over research papers
Python LangGraph Qdrant FastAPI PostgreSQL Docker
- RAG pipeline with section-aware chunking, Qdrant vector search, and paper- and page-level citations
- Hybrid dense + BM25 retrieval with cross-encoder reranking: Recall@5 0.48 → 0.66, MRR 0.30 → 0.45
- LangGraph agent router for evidence Q&A vs. research-gap analysis that routed all 52 test queries correctly
Python PyTorch Hugging Face CodeT5 FastAPI
- End-to-end NLP training pipeline on CodeSearchNet
- Compares a Seq2Seq LSTM, CodeT5 and a decoder-only Transformer for code generation
- FastAPI endpoints return top-k completions for IDE-style autocomplete
Python PyTorch ResNet YOLOv5
- Image-based pomegranate disease detection module built with computer vision
- Benchmarked ResNet classification against YOLOv5 detection to choose the approach
Focus areas: RAG · Embeddings · Hybrid Search · Reranking · Transformers · NLP · Computer Vision
- Semi-Finalist, Flipkart GRID 7.0 (national software engineering competition)
- JPMorgan Chase Code for Good 2025, Hyderabad: built React/Tailwind interfaces for a nonprofit's solution
- Merit Scholarship, Thapar Institute of Engineering and Technology
