GigaInvestors
What 83 widely followed investors own, quarter by quarter from their 13F filings, drawn as treemaps with pencil-sketch portraits.
https://gigainvestors.com
/ — every investor, tile size = portfolio value, scrub back to 2006/BRK — one investor's positions; green = buying, red = selling, scaled by the size of the move/s/NVDA — who holds a stock, since when, and whether they're buying or selling, with shares held stacked by investor over time- Press
/ to search investors, firms, tickers, companies
/newsletter — one letter a quarter, with its archive and open rates- Markdown twins: request any investor, stock, letter or trust page with
Accept: text/markdown (see /llms.txt)
The quarterly letter
When about 80% of the 83 have filed, a GitHub Action builds and sends one issue. lib/newsletter/issue.ts derives the facts (every move sized against that investor's own book, who else did the same, what the price did, how long a position was held); write.ts hands those facts to a model and then refuses any number, ticker or word that cannot be traced back to them; hero.ts draws the lead investor's book as a treemap that doubles as the social card. Nothing is invented, and the build fails rather than sending prose it cannot verify.
npx tsx scripts/newsletter/build-issue.ts # newest quarter, reuses stored prose
npx tsx scripts/newsletter/build-issue.ts "2026 Q1" --rewrite
npx tsx scripts/newsletter/send-issue.ts 2026-q2 --test [email protected]
npx tsx scripts/newsletter/check-send.ts # did the last send actually work
Data
Quarterly SEC 13F filings as compiled by dataroma.com. Values are quarter-end reported values; prices are value ÷ shares, split-adjusted. 13F covers US-listed long positions only and lags quarter end by up to 45 days. Not investment advice.
The dataset lives in the repo (data/store/) and the whole site is static: no runtime data fetching, no accounts, no tracking. The only server code is the handful of newsletter routes under app/api/.
Run it
npm install
npm run sync -- --full # rebuild data/store from dataroma (change-driven without --full)
npm run faces # regenerate face assets from assets/sketches/*.png (not in repo)
npm run dev
npm test # unit tests; `npx playwright test` for e2e against `npm start`
Portraits are pencil sketches generated from public photographs in a single house style. Norbert Lou has no public photograph anywhere, so he is drawn as a mystery man.
