Kyle Tran
Underwriting and automation · Los Angeles
USC B.S. Real Estate Development, Aug 2026
An agent I built reads 412 sources. I edit what survives. Two to four issues a week, mornings Pacific time.
Free · Every claim sourced
412 feeds swept before each issue: RSS, APIs, filings.
Embeddings dedupe roughly 2,100 items down to ~220 distinct stories.
CRE-relevance weighting ranks stories; the top 22 survive to draft.
Claude drafts all three sections with citations back to source.
I read every line, cut what does not hold up, and send it myself.
The whole pipeline runs on a scheduled GitHub Action: async Python fetches and dedupes the week's sources, embeddings land in Postgres via pgvector for clustering, and Claude drafts each section with citations back to source. I read every line before it ships.
July 23, 2026 · 15 stories