I'm an engineer at Bland, where I built our in-house media server and our omnichannel product. Before that I was the first hire at Open Ledger, an embedded accounting API that Collective acquired, and co-founder and CTO of Boondoggle AI.
On the side I build Langbot and do independent research on how language models simulate people.
I gave six models from four developers the personality scores of 1,190 people from the Human Connectome Project and asked them to predict each person's questionnaire answers and measured task performance. The models came close to a trained model on questionnaires but barely predicted behavior. Persona simulations built from self-report reproduce how people describe themselves, not what they do.
QuestionnairesMeasured behaviorTrained modelPearson r between prediction and the real person. Opus 5 ran on 600 people, the rest on 1,190.
Built our in-house media server, which relays RTP audio for SIP calls. It replaced our third-party provider, let us connect directly to carriers, and cut our cost per call by 67%.
Built our omnichannel product: the same agent over voice, SMS, RCS and iMessage, with memory shared across channels. Messaging usage among enterprise customers grew 10x.
Testing and monitoring for trading bots on Kalshi and Polymarket. It records every market in real time and replays order books with real fees and latency. Built solo with Claude Code in six weeks. On election night it recorded 10.3 million trades and transcribed 1.1 million words of live TV on one clock.
Open LedgerFounding engineer, first hire2025
Built the full application: transaction sync, categorization, double-entry books on TigerBeetle and financial reports, shipped as an embeddable library and API. HF0 Residency S25. Acquired by Collective.
Boondoggle AICo-founder, CTO2023–2024
An AI agent that kept sales teams' CRMs up to date from their email and meetings. Backed by General Catalyst and Soma Capital.
Boeing DefenseAutomation software engineer2022–2023
Tools that used language models to draft technical specifications for military aircraft.
Projects
Shakedown2026
Regression tests for robot policies in simulation. With 200 ms of control delay, skipping stale actions raised a public ACT policy from 24% to 80% success.
Splatt2026
Phone scans become walkable Gaussian-splat worlds, with a timeline of every scan of a place.
MachineCI2026
Tests robot-fleet software against independent implementations of the protocols it speaks.
PokéTax2025
A Pokémon-style tax-filing game built on Open Ledger's API. Covered by TechCrunch.