I like finding the understandable system inside the complicated one.

From applications to the systems beneath them

My background crosses product engineering, backend services, cloud infrastructure, machine learning, and AI. That range matters less as a list of technologies than as the ability to move between layers of the same problem.

Over time, my work has shifted from building applications toward building the runtimes, tools, and context infrastructure that make increasingly capable software useful. Term2 explores agent execution and capability boundaries. ChatForge brings those ideas into a complete AI product. code-rag focuses on the context agents need, while easyocr.js carries a computer-vision pipeline into JavaScript runtimes.

How I approach engineering

I care about architecture because it determines how much of a system a person has to hold in their head. My usual loop is straightforward:

1. Understand the existing system and its real constraints.
2. Find the smallest useful abstraction and give it a clear boundary.
3. Build the complete path, not an isolated demo.
4. Verify the behavior that matters in a real workflow.

This is not minimalism for its own sake. Complicated problems still demand sophisticated systems; the goal is to keep that complexity deliberate, contained, and explainable.

Across the stack

I’m comfortable following a problem from model and inference through agent runtime, backend, API, infrastructure, and interface. That breadth helps when a failure sits between conventional ownership boundaries—which is often where the interesting engineering work begins.