Personal Tutor

Overview

Built a sophisticated RAG-based conversational tutoring system using Pinecone vector search for semantic document retrieval. The system processes uploaded documents, chunks them optimally, creates embeddings, and retrieves the most relevant context to ground GPT-4 responses — dramatically reducing hallucinations. Achieved ~1.8s average end-to-end response latency for documents up to 50 pages.

Key Highlights

• Pinecone vector search for semantic retrieval
• 1.8s avg response latency for 50-page docs
• Reduced LLM hallucinations via structured context
• Optimized chunk size & embedding strategy