Organizational Knowledge Graph
Turn thousands of disconnected company documents into an explorable knowledge graph — making implicit organizational knowledge explicit and navigable.
Why this exists
Most organizations accumulate knowledge in documents, wikis, Slack threads, and emails that nobody can actually navigate. Search returns documents; it doesn’t answer questions or reveal relationships.
A knowledge graph makes the hidden structure of organizational knowledge explicit — turning documents into nodes and relationships into edges.
Problem
When a new employee joins or a senior engineer leaves, critical context is lost because it was never formalized. Existing document search is keyword-based, not relationship-aware.
What was built
An end-to-end pipeline that:
- Ingests raw documents (PDFs, Confluence, Notion)
- Extracts entities and relationships using LLM-based structured extraction
- Resolves entities across documents (entity resolution)
- Stores the result as a property graph
- Answers questions by traversing the graph + generating grounded responses
Results
- Processed 2,000+ internal documents
- Reduced context lookup time significantly compared to keyword search
- Graph topology revealed organizational silos not visible from org charts
Technical architecture
Documents → Chunking → Entity Extraction (LLM) → Entity Resolution → Graph Store
↓
GraphRAG QueryEntity resolution is the hard part. See the Embedding Collisions investigation for the specific failure mode that required graph-topology-based disambiguation.