At the Dept. of Computer Science and Engineering, IIT Indore, I designed and built a two-phase Graph-RAG system for Indian parliamentary policy analysis. It combines DuckDB, ChromaDB, and Neo4j to ground language-model outputs in verified parliamentary records.
I also engineered a two-stage LoRA fine-tuning pipeline that produced sansad-v2 from a Mistral-7B base, then quantized it for local Ollama deployment. In an automated evaluation across 150 questions, sansad-v2 achieved 92.89% Faithfulness and 91.18% Citation Accuracy.
At IIM Udaipur, I led end-to-end analysis of an e-commerce dataset with 3,600+ customer sessions, using correlation analysis and logistic regression to identify conversion drivers. I also examined how engagement patterns related to conversion rates.
My projects include a YOLOv8 license plate detection system and a medical assistant RAG system for answering questions about uploaded PDFs. I work across machine learning, NLP, data analysis, and retrieval-augmented generation, with Python as my main language.
