Most of what mattersis already public.Almost none of itis assembled.
✦ Read the public surface. Assemble the structure underneath.
The intelligence layer
Not an AI that writes investment reports
An autonomous US venture-intelligence engine that discovers early-stage startups, reconstructs founders and companies from public evidence, evaluates IP and competitive advantage, detects hidden signals, models US regulatory and market context, predicts innovation trajectories, and produces evidence-backed angel-investment intelligence.
Scope for v1 is deliberate: US-headquartered or US-incorporated companies. Delaware status, Form D filings, USPTO assignments, federal procurement and state licensing are legible, dated, and adversarially reliable in a way a global corpus is not.
Five chapters, each its own route.
The system is too large to read as one scroll. Each chapter stands alone, carries its own design system, and links on to the next.
- 01Evidence before opinionDiscovery, entity resolution and the typed evidence graph — plus the model routing that decides which tier ever sees a token.
- 02Where the network already talksThe US Context Pack, the twelve densest clusters, the IP questions that break diligence, and how claims are held against time.
- 03Four classes of derived signalFounder, technical, commercial and network alpha; the trajectory model; and the same company read two ways.
- 04What an investor actually receivesThe evidence-backed memo, its unresolved questions, and the monitor that keeps re-reading after the memo ships.
- 05A running process, not a slideLive row counts, cadence, latency distributions and recent passes, read straight from the store over SSE.