A backend engineer at Postype shares how they built an internal RAG-based chatbot to help employees query up-to-date system specs directly from source code. The core insight: 'code is always the latest spec.' The post covers the full AWS Bedrock Knowledge Base setup including chunking strategy (method-level Java parsing), metadata filtering to narrow search scope from 4,400 to ~50 documents, hybrid search (semantic + keyword via RRF), and why Rerank was ultimately rejected for code retrieval. A CI/CD pipeline triggers automatic KB sync on every production deployment, keeping the knowledge base current within 5 minutes.