A full-stack ChatOps assistant that indexes Markdown, chunks content by sections, combines textual search with pgvector-ready storage and composes structured answers without a mandatory LLM dependency.
- FastAPI
- React
- PostgreSQL
- pgvector
- Celery
- Redis
- Docker
Context and problem #
Repeated incidents require consistent answers. Markdown runbooks concentrate operational knowledge, but need to be searchable and actionable.
My contribution #
The project demonstrates backend engineering, a deterministic RAG pipeline, support frontend and adapters for operational channels.
Architecture #
Runbooks with YAML frontmatter are parsed, chunked by heading, indexed and queried through hybrid search; the answer exposes sources and confidence in the React dashboard.
- ✓ Markdown parser
- ✓ PostgreSQL + pgvector
- ✓ Answer composer
- ✓ Slack/Discord/webhook adapters
Evaluation and quality #
Quality appears through traceability: each answer must carry sources, verifiable steps and feedback to measure hits and misses.
Limitations #
The assistant supports operations, but should not execute critical commands without human review and permission controls.
Next steps #
Add authentication, permissions, real Slack/Teams integrations, observability and optional LLM support.
Test the product and inspect the code
The live demo runs the core flow with safe local data. The public repository preserves the complete backend, tests, infrastructure and technical documentation.