Why we built our own LLM
The build-vs-buy decision at the model layer, told from the inside. What it takes for a small team to train and ship a domain-specific LLM (data flywheels, evaluation, serving economics) and how that calculus looks today.
Speaking
Talks
I occasionally give talks, mostly about things I've built. Previously on stage at API:WORLD.
Things I talk about
Are human conversations special?
What attention patterns reveal about how language models handle real human dialogue, why models that master code and math still stumble on conversation, and what that means for anyone building voice agents.
AI systems in production: what breaks and why
Eight years of shipping AI, from real-time pipelines to LLM-powered agents at enterprise scale. Latency budgets, evals that predict real quality, and the last 10% that turns a demo into a product.
Building the teams that build AI
Running AI engineering through a platform shift, an acquisition, and into enterprise scale, and how the job differs between a twelve-person startup and a company that bets revenue on your models.
Want to talk about any of this, on a stage, a podcast, or over coffee? Email me.