The open lab

Less abstract.
More hands-on.

Open-source experiments and personal applications, each built around a specific problem. Explore what they do and why I built them.

AI & engineering projects

Browser app / Public beta

Agent Failure Recipes ↗

Reproduce a failure, make one repair, and keep the test. Three executable TypeScript recipes for the software failures around tool-using agents.

Open the app ↗View source ↗

Agent Failure Recipes comparing two committed tickets in the broken run with one ticket after the idempotency repair, alongside an execution trace and authoritative state.

Browser app / Public beta

Task Cost Calculator ↗

Compare the cost of acceptable AI outcomes, including failed attempts, tools, infrastructure, and human review.

Open the app ↗

Browser app / Public beta

Agent Explainer ↗

Watch an agent fail, inspect the cause, apply one repair, and replay. An interactive lab for understanding what happens between a request and its result.

Open the app ↗View source ↗

Agent Explainer showing its experiment picker and separate panels for context, activity, and actual world state.

Personal projects

Browser app / Personal project

Veer Reading Quest ↗

Read. Picture it. Connect the clues. Prove your answer. A calm reading-practice app I built for my son Veer, with two learning levels and parent review.

Open the app ↗

The reading method: read, picture the story, find who did what, connect the clues, draw, and prove the answer with words from the passage.