Async on-prem LLM-powered structured information extraction microservice
- nvidia/cuda:12.4 runtime base matches the deploy host's driver stack (immich-ml / monitoring use the same pattern). - python3.12 via deadsnakes (Ubuntu 22.04 ships 3.10 only). - System deps: libmagic1 (python-magic), libgl1/libglib2 (PIL + PyMuPDF headless), curl (post-receive /healthz probe), ca-certs (httpx TLS). - uv sync --frozen --no-dev --extra ocr installs prod + Surya/torch; dev tooling stays out of the image. - CMD runs `alembic upgrade head && uvicorn ix.app:create_app` — idempotent. - Compose: single service, port 8994, GPU reservation mirroring immich-ml, labels for monitoring dashboard auto-discovery + backup opt-in. - host.docker.internal:host-gateway lets ix reach the host's Ollama and postgis containers (same pattern mammon uses). Task 5.1 of MVP plan. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|---|---|---|
| .forgejo/workflows | ||
| alembic | ||
| docs | ||
| scripts | ||
| src/ix | ||
| tests | ||
| .env.example | ||
| .gitignore | ||
| .python-version | ||
| AGENTS.md | ||
| alembic.ini | ||
| docker-compose.yml | ||
| Dockerfile | ||
| pyproject.toml | ||
| README.md | ||
| uv.lock | ||
InfoXtractor (ix)
Async, on-prem, LLM-powered structured information extraction microservice.
Given a document (PDF, image, text) and a named use case, ix returns a structured JSON result whose shape matches the use-case schema — together with per-field provenance (OCR segment IDs, bounding boxes, cross-OCR agreement flags) that let the caller decide how much to trust each extracted value.
Status: design phase. Implementation about to start.
- Full reference spec:
docs/spec-core-pipeline.md(aspirational; MVP is a strict subset) - MVP design:
docs/superpowers/specs/2026-04-18-ix-mvp-design.md - Agent / development notes:
AGENTS.md
Principles
- On-prem always. LLM = Ollama, OCR = local engines (Surya first). No OpenAI / Anthropic / Azure / AWS / cloud.
- Grounded extraction, not DB truth. ix returns best-effort fields + provenance; the caller decides what to trust.
- Transport-agnostic pipeline core. REST + Postgres-queue adapters in parallel on one job store.