Aura Neural Search.
Experimental semantic search engine leveraging vector embeddings for natural language querying.
Python
LangChain
Pinecone
Next.js
Overview
Aura explores the intersection of LLMs and structured data. It allows users to query complex databases using natural language, converting intent into SQL/Cypher queries and summarizing results with cited sources.
Technical Challenge
Reducing hallucination in SQL generation. I implemented a 'Schema-Aware' RAG pipeline that retrieves only relevant table schemas and sample rows before prompting the LLM, combined with a deterministic SQL validator that dry-runs queries before execution.