Tech
How we build, ship and keep it running. Repositories, architecture references, environment access and the on-call rotation — all in one place.
Tech resources
The daily reference
Repositories
The Acuminator GitHub org. Single-instance product: pushing to main on
acuminate-rpl deploys the live pilot. No UAT buffer.
Architecture
4-app monorepo: FastAPI backend, React/Vite frontend, standalone voice service, and the
ace_rpl Python library for parsing & generation. See AGENTS.md.
Environments & Deploys
Railway hosts the live services (backend-web, worker, frontend, voice). A parallel Sydney Cloud
Run stack on GCP acuminate-4c579 is staged for cutover.
Generation Pipeline
Celery task parses TGA units → LLM-primary generation (fail-loud, no silent fallback) → Word/JSON outputs, cached per unit·qual·version.
rpl-backend/app/tasks/generation.pyStandards & Testing
TypeScript strict on the frontend; ruff + pytest on the backend/library; Playwright e2e. CI runs
on every push/PR to main.
AI / Inference
LLM providers behind ace_rpl: Anthropic (default Sonnet), OpenAI, Gemini. Gap
analysis, risk profiling and the evidence pipeline live here.