member of technical staff @ taste labs · washu cs
building rl environments for frontier ai labs. building slop bench, a forthcoming public benchmark for how well frontier models produce intricate, high-quality web design.
languages: typescript · go · java · python · ruby · c++
frameworks: next.js · react · spring boot · fastapi · rails
infra: docker · kubernetes · traefik · aws · gcp · github actions
data: postgres · supabase · redis · mongodb · sqlite
ai/ml: rl environments · harness engineering · mcp · agent evaluation · inference hostingpersonal site and blog. portfolio of projects, writing on engineering and trading systems.
stack: plain html · tailwind v4 · markdown posts · vanilla js
hosted: cloudflare pages
status: liveautonomous options trading. one claude agent ("claudia") ran a single real brokerage account. three sessions a trading day it read the book, the tape, and the news, then traded options (calls and puts), sized in code so no single contract or name could sink the account. benchmarked against buy-and-hold spy, with every move public, tool call by tool call.
stack: go · next.js · postgres · traefik
data: schwab trader api · anthropic · resend
features: live quotes · daily session transcripts · eod p&l vs spy · recap email
status: discontinued · repo archived for reference