Pre-Final Year · IIT Madras · Quant Finance + Materials Engineering
LinkedIn · Email
Pre-final year student at IIT Madras interested in AI/ML and full-stack development — particularly how GenAI integrations change the way users interact with applications. Actively exploring RAG, NLP, agentic systems, and full-stack development through projects and internships.
RAG & GenAI Intern · Tecnod8.ai · Sep–Nov 2025
- Built production RAG pipelines (hierarchical chunking, HyDE) for enterprise document retrieval.
- Developed a multilingual layout extractor combining DocLayout-YOLO, PaddleOCR, and VLM refinement.
- Ranked top 13 / 150+ teams in the NCIIPC Grand Challenge RAG track.
AI & Software Engineering Intern · Genius Labs · May–Jul 2025
- Built the front-end UI of an AI-powered exam prep app using Next.js.
- Designed a test-management framework integrating generation, participation, and evaluation workflows.
| Competition | Result |
|---|---|
| IMC Prosperity 2025 — Algorithmic Trading | #69 globally / #10 nationally among ~9K teams |
| WorldQuant IQC 2025 | Round 2 qualifier · #11,597 / 88,000+ |
| NCIIPC Grand Challenge — RAG Track | Top 13 / 150+ teams |
Movie Recommender — Full-stack recommendation platform
Next.js FastAPI MongoDB JWT Sentence-Transformers
- Semantic matching on 10K+ movies via Sentence-BERT; SVD collaborative filtering achieving 78.8% Precision@10 at <100ms latency.
LiquidCuts — Financial news & liquidity tracker
Next.js MongoDB Google Search API yFinance
- Evaluates stock price reactions to 100+ event-driven articles; classifies layoffs and legal events via keyword extraction.
Financial News Sentiment Analyzer — Sentiment classification pipeline
Python PyTorch Scikit-learn Streamlit
- Benchmarked TF-IDF, Word2Vec, LSTM, FinBERT; improved accuracy from 51.7% → 70.5%. Deployed on Streamlit.
Retrieval Techniques Benchmark — RAG retrieval evaluation
Python BM25 Dense Retrieval RRF Cross-Encoder
- Improved context recall from 72.8% → 81.4% on BEIR SciFact (5K docs) via hybrid retrieval + reranking at 51ms latency.
LocalRAG — Offline PDF query engine
Python Ollama ChromaDB LangChain SentenceTransformer
- End-to-end local RAG pipeline — ingestion, chunking, embeddings, retrieval, and automated evaluation. Zero cloud dependency.
Document Layout Detection (WIP) — Layout detection for complex PDFs
Python DocLayout-YOLO PaddleOCR VLMs OpenCV
- Detects layout regions on PDFs; VLM refinement step corrects misclassified boxes using OCR context and full-page images.
MCP Generator (WIP) — Agentic MCP tool generator
Python LangGraph LangChain Node.js
- 8-node LangGraph pipeline that converts natural language into executable Rocket.Chat MCP tool modules.
Python JavaScript C C++ · PyTorch Scikit-learn HuggingFace Pandas NumPy · RAG LangChain LangGraph NLP · Next.js React FastAPI MongoDB Node.js · Git Streamlit Docker
- Advanced Learning Algorithms · Regression & Classification — DeepLearning.AI
- Probability & Statistics · Data Structures & Algorithms