Ask “who worked on the Acme migration?” and a pile of documents won’t answer it. You need the entities behind the documents — people, teams, accounts, engagements — and the relationships that connect them.
That’s the entity layer: a canonical map of the organization that every chunk of indexed text points into. A passage isn’t just words; it’s words about specific people and specific work.
Why the graph matters
Resolving a question to entities lets an answer be both precise and traceable. “The team on Acme” becomes a concrete set of people, each linked to the evidence of their involvement.
| Question | Resolves to |
|---|---|
| Who owns this account? | A person entity + the source that says so |
| What did this team ship? | Engagements linked to that team |
| Where is this project discussed? | Chunks tagged with the project entity |
The map is what turns a search engine into a memory. It knows not just what was said, but who it was about — and can hand you the receipts either way.