The Ramco Aviation
Brain
A queryable, source-truth Brain over a real aviation MRO ERP. Six layers — data, graph, wiki, SDK, runtime, journeys — all grounded in the operator's own data and documentation.
The Brain stack
Source truth from raw Ramco data → through to two journey-driven demos.
ramco_aviation.sqlite — 605 MB, 152 tables, real Ramco MRO data filtered to top-20 aircraft
92 inferred SQL joins between 124 entity types — traversable from any aircraft
326 markdown pages with [[wikilinks]], YAML frontmatter, dense cross-references
2 executable workflows with chat-enabled decision points + API hooks to a mock Ramco runtime
Two journeys
Each is a sequence of decision-making points. The chat is open at every step.
Crew reports an intermittent hydraulic warning. Walk the maintenance supervisor through 6 decision points: raise discrepancy → investigate → defer or fix → execute → sign off → COM. Real Ramco APIs hit the runtime at every commit.
The component side. The pump removed in Journey A enters the shop. 5 decision points: open SWO → route → inspect (REPAIR/NFF/BER) → consume parts + test → conformity cert. Loops on test fail.
Did it pass the test?
100 hand-authored ground-truth questions, programmatically scored against the live deployment.
Every reply scored automatically against expected substrings, numeric tolerances, and set-membership checks. Categories: schema, live DB counts, ontology, relationships, rules, journey, domain knowledge, cross-domain reasoning. The 5 failures are real and documented.