Case Study • Document Intelligence
A local-first document intelligence platform that ingests unstructured pitch decks, parses complex layouts and generates structured, evidence-backed investment scorecards.
Investment teams get more decks than they can read. Most tools that promise to help require uploading confidential decks to someone else's cloud, which many funds simply can't do.
And a score alone doesn't help. An analyst has to be able to check where a market-size claim came from, or they'll end up reading the deck anyway.
An extracted fact without a link to its slide is just a rumor.
Deterministic extraction first, models where they help, and evidence for everything.
The entire pipeline works with a local model, so sensitive documents stay private.
Each extracted fact carries a reference to its slide, so verifying takes one click.
Scorecards measure a startup against the fund's own configured investment thesis.
Five stages, each resumable and each leaving an inspectable artifact.
Strict data privacy comes from where the system runs, not from a promise in a policy.
Regression fixtures and model-call cost and latency tracking keep changes measurable.
Docker deployment, upload and auth controls, retention, backups and a runbook are all in the box.
