Intelligent Query Routing and Answering System

Led the design and end-to-end development of a hybrid query routing system as a proof of concept (POC) for a major US-based auto insurance client, which directly contributed to securing the project.

Built a LangChain-powered data pipeline using Qwen-QWQ-32B on Groq for dynamic SQL query generation and retrieval from Postgres, enabling accurate, LLM-generated responses to user queries.

Developed a RAG workflow with ChromaDB and Google’s Text Embedding 004 for vector-based retrieval, integrating UnstructuredLoader to process PDF policy documents and deliver context-rich answers to policy-related questions.

Reduced hallucinations and improved trust in AI outputs by integrating a hallucination grading and reranking mechanism using cross-encoders, significantly enhancing response quality during client demos.

Engineered an intelligent hybrid router that dynamically split user queries across relational and vector stores, optimizing for speed and accuracy; showcased as a key differentiator in the sales demo.

Collaborated with a sales-aligned SME for domain guidance and worked alongside a frontend developer for UI integration, ensuring a complete and demo-ready POC.