Work · RAG · In use across team workspaces
Knowledge Ingestion Pipeline
Watches for new documents and turns them into knowledge you can search by meaning or by relationship.
PythonClaude APIChromaDBNetworkX
context
Built at work for the solution architecture team
The problem
Playbooks, API docs and client notes piled up in folders. Finding the one answer meant reading everything again.
What I built
- An event-driven watcher that picks up new files the moment they land.
- Claude extracts structured knowledge into typed schemas, with retries when a call fails.
- Each source is indexed three ways: linked Obsidian notes, a NetworkX knowledge graph and ChromaDB vector search.
- One command to digest a folder and one to query it.
How it works
A folder watcher queues new documents automatically.
Impact
- Team knowledge became searchable by meaning, not just by filename.
- Shared across several project workspaces.
What I learned
Structure beats raw chunks. Extracting entities and links first made retrieval far more reliable than embedding whole pages.