Agentic Engineering Memory System

Most AI developer tools are simple "chat with code" wrappers. The Agentic Engineering Memory System goes further by turning fragmented development activity into durable organizational knowledge. It captures the "why" and "who" behind the code: architectural decisions, historical timelines, past incidents, ownership, and unresolved questions. How it works: Asynchronous Ingestion: Connect a GitHub repository, and the system starts background ingestion of PRs, issues, comments, and commits. Layered Memory Extraction: Raw sources are treated as evidence. A background pipeline autonomously extracts structured memories, mapping what decision was made, why, and who was involved. We use a strict layered approach (keeping raw evidence, derived memory, and retrieval context separate) and apply confidence filtering so weak AI guesses do not pollute memory. Dynamic Retrieval: Engineers ask questions via a Streamlit interface (e.g., "Why was Redis removed?"). The LangGraph orchestration layer plans the retrieval path dynamically. It pulls memories using semantic search, graph relationships, and timeline lookups, then uses IBM watsonx.ai to synthesize an answer with exact source traces. The Role of IBM Bob: Building an autonomous, multi-agent system is complex. We utilized the IBM Bob IDE as our build copilot and proof-of-work layer. Bob assisted in scaffolding our LangGraph architecture and asynchronous pipelines. The exported IBM Bob task session report in our repository serves as our build ledger.