Nityashree S

I built an LLM-powered Advanced RAG Document Assistant for research paper comprehension, improving retrieval relevance through hybrid BM25 and FAISS retrieval, query rewriting, and cross-encoder re-ranking.

I designed its LangGraph workflow with guardrails and delivered a user-facing application using FastAPI and Streamlit. My work spans Python, LangChain, LangGraph, Gemini 2.5 Flash, FAISS, ChromaDB, and agentic AI.

At Nuclear Power Corporation of India Limited’s Kaiga Site, I translated safety-critical requirements into an ESP32-based automatic shutdown prototype. I also studied COIS and PDCS workflows and collaborated with engineers to validate optical transducer performance before deployment.

I'm also developing G-ID, a sensor-fusion system for real-time doctor–patient communication through Tamil Sign Language translation. As a co-author of an accepted IEEE/CVF CVPR Multimodal Workshops 2026 paper, I helped benchmark six architectures and co-design a 30-signer dataset that showed a simpler sensor-only model outperformed multimodal fusion.