Skip to content

00. Evidence Map

Repository inspected

Item Value
Remote https://repo.aibott.org/drt/ai-use-cases (private)
Local checkout repos/ai-use-cases
Inspected on 2026-07-09
HEAD at clone 09411cb43aa32325518d9b324d91c2025d035965

Areas inspected

Area Paths Depth
Manifests backend/pyproject.toml, backend/uv.lock, frontend/package.json, frontend/pnpm-lock.yaml Full
App entry backend/app/main.py, config.py, dependencies.py, middleware.py Full
Ingestion / jobs API core/ingestion/router.py, core/extraction/router.py Full
Pipeline workers/tasks/pipeline_tasks.py, async_runner.py, celery_app.py Deep skim
OCR core/ocr/engine.py, vlm/, sarvam/, layout/ Entrypoints + factory
UC1 factsheet uc1_factsheet/router.py, service/phase1.py, phase2.py, factsheet.py, models.py Deep skim
Data db/models.py, db/session.py Full
Frontend App.tsx, api/*, pages, ProvenanceTooltip.tsx Full / deep skim
Infra infra/docker-compose.yml, docker-compose-full.yml, Makefile, .env.example Full
Docs / plans README.md, docs/**, plans/prd_*.md, phase plans Skim + PRD/use-case plans
Tests backend/tests/, backend/tests/unit/ Listing + skim

Intentionally skipped

  • Lockfile contents beyond dependency names
  • Generated volumes/, logs/, .venv, node_modules
  • Live OCR/LLM/Celery runs (no stack started during inspection)
  • Line-by-line read of every shadcn primitive under frontend/src/components/ui/
  • Full MkDocs wiki body (structure and key architecture pages sampled)

Commands run during inspection

git clone https://repo.aibott.org/drt/ai-use-cases.git repos/ai-use-cases
find / rg inventory of source tree
# Targeted reads of README, config, models, routers, pipeline, factsheet, frontend, compose

No pytest or Docker stack was executed during dossier authoring (Unknown: live OCR/LLM behavior).

Confidence by subsystem

Subsystem Confidence Notes
Document upload + job API High Routers + models + tests present
Celery OCR→extract→merge pipeline High Large pipeline_tasks.py; stage enums match API
OCR engine factory High OCR_ENGINE wiring clear; providers not live-tested
UC1 factsheet phases High Phase modules + API + unit tests
Frontend operator console High Routes and API clients read
Auth / OIDC High (absence) Helpers unused on routers
Qdrant / Meilisearch High (absence) Config + compose only; no app call sites
Production deploy Low Docs describe CI/K8s; no workflows or Dockerfiles for app

Confirmed / Inferred / Unknown

Confirmed

  • FastAPI app version 2.0.0 with routers under /api/v1 for documents, jobs, factsheet
  • Postgres models for documents, jobs, stages, chunks, artifacts, audit events, factsheets
  • Schema created via Base.metadata.create_all on lifespan (no Alembic migration tree)
  • Celery workers use run_async_in_worker_loop for async DB/LLM
  • UC1 is the only use-case package under backend/app/
  • CORS allow_origins=["*"] with credentials enabled
  • Compose maps Postgres to host 5433; MinIO on 9000/9001

Inferred

  • Intended users are DRT registry/scrutiny operators (from PRD), though UI has no role model
  • README “vector & full-text search” describes target architecture, not current code
  • progress_and_context.md may be stale relative to implemented UC1

Unknown

  • Whether Meilisearch/Qdrant were used on other branches
  • Production SSO issuer and deployment topology
  • Whether MINIO_* vs S3_* env mismatch is intentional dual naming or drift
  • Intended Vite proxy port (8888 vs Makefile 8000)