My Projects

Domain-Specific RAG Chatbot

• Designed a distributed architecture separating vector storage (ChromaDB), embedding generation, and LLM inference into independently deployable services.
• Fine-tuned Mistral 7B using LoRA/QLoRA — achieving higher domain accuracy and reduced hallucination rates compared to the base model.
• Implemented retrieval chains, prompt templates, and evaluation loops with LangChain and e5-large-v2 embeddings for semantic retrieval.
• Stack: Python, LangChain, HuggingFace Transformers, ChromaDB, FastAPI