Blog
Posts and writing on engineering, AI, and building.
Lessons from Eight Years Building Startups
Traction isn't product-market fit. Timing matters as much as execution. And the cost is real: relationships, health, money, mental health.
Build Your Own LLM Eval Harness vs. Off-the-Shelf Observability
The real tradeoff between rolling your own eval pipeline and using a platform—and when each one actually pays off.
Building Untrace and Acquiring to OpenRouter
Why I built an LLM observability SDK, how it worked, and what happened when OpenRouter acquired it.
What Actually Breaks When You Ship an LLM Feature to Production
Latency variance, guardrails, retries, cost explosions, and prompt drift—the things nobody tells you until you're on fire at 2am.
The Missing Developer Tool: An LLM Local Debugger
Why debugging LLM systems sucks, what exists today, and what a real debugger would look like.
The Ideal AI SDK
What a great AI SDK would do: streaming, structured outputs, tool calling, evals, and observability—without the pain of today's options.
Why Prompt Engineering Is the Wrong Abstraction
Raw prompt tweaking doesn't scale. Structured prompts, DSLs like BAML, typed outputs, and eval pipelines are the right abstraction.
What I Learned Building 20 AI Prototypes in a Year
Fast feedback loops, when to use which model, and why most prototypes should stay prototypes—until one shouldn't.
The Hidden Architecture Behind Good AI Products
The stack most people miss: user input → guardrails → prompt assembly → model routing → streaming → post-processing → observability.
The New Startup Stack for AI Products
A practical stack for shipping AI in 2025: Next.js, OpenRouter, Langfuse, pgvector, and where to run workers—with tradeoffs.
How to Stream Structured JSON from an LLM
Get typed JSON from the model without waiting for the full response—streaming, parsing, and handling partial chunks.
The Real Cost of Running LLMs in Production
Tokens, latency, infra, observability, and evals—what actually shows up on the bill and how to stay in control.
The AI Application Layer Is Still Wide Open
Why the best AI products aren't the ones with the biggest models—they're the ones that nail the workflow, distribution, and data.