How I Designed an Enterprise RAG System Using AWS Bedrock, Pinecone & Neo4j

AI Agent Building Project Overview In my role as an AI Solutions Engineer at Prescott Data, I built an enterprise-grade document Q&A chatbot for a client whose internal policy documents were difficult to search, inconsistent in structure, and costly for teams to navigate manually. Employees struggled to find accurate answers quickly, especially across long, multi-column PDFs with cross-referenced policies. The solution was a retrieval-augmented generation (RAG) system that combines vector similarity search (Pinecone) with knowledge graph reasoning (Neo4j) to deliver accurate, context-aware responses. Tech Stack: Backend: AWS Lambda (Python), AWS Bedrock (Titan Embeddings & DeepSeek-R1) Databases: Pinecone (vector store), Neo4j (knowledge graph) Document Processing: AWS Textract with LAYOUT feature for multi-column PDFs Frontend: React with Markdown rendering Session Management: DynamoDB for conversation history System Architecture