I'm principal architect and lead engineer of Darcy's data management platform. It replaces spreadsheets with one system for designing geothermal wells, approving them, and recording what crews actually install.
I took it from an early graph prototype to production in under a year. The back end is FastAPI and Neo4j, the office works in an Angular web app, and field crews get a native iPad app that works offline.
Each well is a versioned graph, so every design revision and every as-built record is kept. The form a crew fills out in the field is built from that well's design, so they see only what applies to it.
AI agents write most of the code under the process I designed: a written spec for every feature, specialist agents building in parallel, reviewer agents that can't edit code, and nothing merged without evidence I can check.