Scout
Ten ecosystems watched. New companies found, researched against the fund's thesis, and ranked into one digest a week.
What it replaces.
Finding new companies by hand means opening every source, reading through it, and looking up anything promising, before a single company has been decided on. It takes hours out of every week, and it starts over on Monday.
Then again next Monday.
Monitor, identify, brief.
Scout checks every source on a schedule, spots the companies it has not seen before, researches each one on the open web, and writes it up in the same format every time.
One email a week.
All that monitoring becomes one quiet email on Monday morning: the new companies ranked against the current thesis, the source each one came from, and the ones held back for a human read.
It keeps looking after everyone logs off.
Between runs Scout remembers what every source looked like, which companies it has already researched, and which names the fund has already met. Nothing is researched twice, and a run that dies halfway picks up where it stopped.
How it's built
| Language | Python. httpx and selectolax for static sources, a headless browser for the JavaScript-rendered cohort pages, asyncio for concurrency. |
|---|---|
| Research | An LLM with web search: founders, stage, traction, funding signals, and the competitive set the company is stepping into. |
| Memory | SQLite. Last-seen fingerprints, the canonical seen-set, and the research cache. A run killed halfway resumes without re-scraping or re-researching. |
| Output | An HTML email: ranked companies, the source each one came from, and links to continue the research. |
Back to the work
BDR Desk, Private Books, Prospecting Agent, Brix, Paylit, and Scout.