BRAIN: a self-improving knowledge engine

BRAIN is how DDD remembers and reasons. It is a knowledge engine that takes everything we learn, from client work to research, and turns it into structured knowledge that improves itself over time.

The challenge

A consultancy lives or dies on what it knows. Notes in scattered documents go stale, contradict each other, and never get revisited. We wanted the opposite: a knowledge base that actively gets sharper, flags its own contradictions, and can be trusted because every claim carries its evidence.

What we built

BRAIN is a local-first knowledge system backed by an Obsidian vault, exposed to our AI tools through the Model Context Protocol. It ingests almost anything: text, URLs, PDFs, code, and more. It breaks what it ingests into atomic claims, each one tracked with its supporting evidence and a computed verification depth, so a well corroborated fact is visibly different from an unverified one.

A set of autonomous routines run on a schedule to audit claims, hunt for corroboration, surface contradictions, and propose what to research next. A live 3D graph lets us see the whole knowledge network at once. The vault stays the source of truth, so the knowledge is portable and never locked in.

How we built it

BRAIN is built in TypeScript on Node, with the MCP SDK, an Express HTTP layer, SQLite full text search for fast retrieval, and Three.js for the graph view. It is covered by a real test suite, because a knowledge engine you cannot trust is worse than no engine at all.

The result

BRAIN is why our advice compounds instead of resetting with every engagement. It is also a clear demonstration of how we think about systems: evidence first, self correcting, owned by you, and built to get better on its own rather than rot.