Who is getting funded?
Each investment gets a card with the company name, a plain description of what it does, the amount and stage, and a link to its site.
What it does
The email answers 3 questions for you.
What sectors are getting investments?
A sector rollup sits above the cards — total capital deployed, deal count, and a per-sector breakdown compared against the trailing four-week average, so a genuine shift is distinguishable from a noisy week.
Is there a pattern?
One email will track funding across 14 different blogs in order to see if and what patterns are forming on companies getting traction.
How it works
Summarizes 14 different newsletters and blogs
Collects 14 newsletters and blogs
Determines what are new posts from the week before
Reads the all new posts
Extracts important information using Claude API
Dedup
Organizes and finds trends
Email sent
Detecting what's new
The agent keeps a memory of every post URL it has ever seen in data/seen_posts.json. Each run scrapes the 14 index pages, compares against that memory, and only the URLs it has never seen continue down the pipeline.
Publication dates are deliberately not the primary signal. Many of these blogs show no date on the index page, some show relative dates, and posts get backdated or quietly edited. The URL diff is reliable where dates aren't. A parsed date, when available, only serves as a sanity filter to reject posts older than ~60 days.
Reading each post
New posts are fetched and parsed for their full body, then passed through a classifier: is this actually announcing a specific company raising a specific round? Most posts on these blogs aren't — they're essays, market maps, podcast episodes, partner announcements. Those are dropped before extraction.
What survives goes through a regex pass for the amount and stage, then a Claude call for the company description, canonical URL, sector, and co-investors. The two passes cross-check each other; disagreement lowers the confidence flag rather than silently picking one.
The extraction prompt specifically guards against the valuation trap — a post reading "raised $30M Series B at a $300M valuation" must report $30M, not $300M.
Sector trends
Every investment is assigned exactly one sector from a fixed list. This is the whole reason the trend view works: a free-form label would drift between weeks and nothing would be comparable.