input · a curious human

About

the shoggoth · mostly aligned · tap it

I'm a Data Scientist at Meta, working on evals for our internal AI Agent Harness. Before that, most of my time went to figuring out whether our internal AI tools actually helped our sales teams — product funnel analysis, voice-first agent demos, and months-long experiments measuring whether the tooling made anyone more productive.

Outside of Meta, I've gotten ~slightly~ AI-alignment-obsessed. 2025 was for fundamentals: I built a GPT from scratch, red-teamed small models for deceptive alignment, and started building benchmarks to dig into the psychology of these new "alien minds."

2026 has been for shipping: KahneBench ran 6 frontier models through 15 Kahneman-Tversky biases (Opus 4.6 came out least biased, if you're wondering), and in June I shipped ShipFastEvals, my mildly-technical guide to whether your AI product actually works. I'm still writing essays too — the latest argues the frontier labs finally turn a profit right when they run out of compute.

location·San Francisco, CA

learned parameters · skills

Skills

Languages

  • Python
  • SQL
  • R
  • TypeScript

ML / AI

  • PyTorch
  • TensorFlow/Keras
  • XGBoost
  • scikit-learn
  • statsmodels

Data

  • PySpark
  • PostgreSQL
  • AWS Redshift
  • Hive/Presto
  • Tableau

Tools

  • Cursor
  • Claude Code
  • Docker
  • Git
  • Jupyter

output layer

Let's talk

P(we should talk) = 0.97