All projects
Case studyStartup intelligence Screens How it's built
Startup intelligence · Node + MongoDB
Strata
Tracks startups by sector from public data, classifies them with LLMs and builds sector landscapes where every claim cites a source.
465
Companies tracked
136
Form D filings verified
3
Sectors mapped
129
Automated tests
Node.jsExpressMongoDBNext.jsKafkaRedisBullMQLLM APIs
- Problem
- Startup databases are expensive, and their numbers are hard to check. An analyst mapping a sector needs to know which companies belong in it, what they raised and where each number came from.
- Who it's for
- Market research analysts and investors who map a sector, track new rounds and write landscape briefs.
- Design decisions
- An analyst's workbench: a market map per sector, company profiles with a funding timeline, URL-shareable search and filters, a ⌘K palette, keyboard-driven review, watchlists and CSV export. Light and dark themes, responsive, with automated accessibility scans.
- Technical decisions
- Express 5 and TypeScript on MongoDB, with analytics written as aggregation pipelines and cached in Redis. Connectors (YC directory, SEC EDGAR, RSS, GitHub) publish to Redpanda (Kafka); a processor upserts on natural keys, and BullMQ runs classification, alerts and briefs. OpenAI, Anthropic and Gemini sit behind one provider interface. Next.js proxies the API so sessions and keys stay server-side.
- The hard problem
- Linking SEC filings to the right company. A Form D is filed by a legal entity, not a brand, so the matcher only accepts a filing when the names match, the filer is not a fund vehicle and the dates are plausible. It kept 136 of 234 candidates, each rejection with a written reason, because a wrong funding round is worse than a missing one.
- Result
- A full Docker stack seeded with 465 real YC companies across Legal AI, Climate Tech and Payments, a written Legal AI landscape, and briefs that drop any claim whose numbers are not in the cited source. 97 API tests and 32 Playwright tests run in CI against the whole stack.
Dashboard: KPIs, quarterly funding by sector and the latest rounds from primary sources. 
Review queue: low-confidence labels, triaged from the keyboard. 
Analyst brief: every claim footnoted to a stored source. 
Company profile: Form D funding timeline, sources and classification history.
Data
MongoDB aggregations, Redis cache
Pipeline
Kafka facts + BullMQ jobs, replay-safe
AI
3 LLM providers, review queue, cited briefs
Next project
HiveState