Open source
The code behind the ideas. I build in the open: benchmarks that measure what AI evaluations miss, safety tooling for the agent era, and small tools that prove useful software does not need your data.
01 · Open source
A benchmark for AI agents on fragmented market intelligence
39 tasks where the signal is scattered across sources and time: the agent has to piece together spatial-temporal market intelligence from fragments. Corpus and scorer frozen at v1.0. Built to measure what AI evaluations usually miss.
02 · Open source
Static supply-chain checker for agent skills
A deterministic, offline checker for agent-skill supply-chain risk: point it at a skill directory and get a structured risk report. No LLM calls, no network, no guesswork. Launched September 2026.
03 · Open source
Headless-browser QA that can only reach localhost
An MCP server with 12 tools for agent-driven QA of local dev servers: navigate, snapshot, screenshot, click, fill, console and network checks. Localhost is enforced at two layers, so agents physically cannot wander onto the public web.
04 · Open source
Client-side price comparison, no backend
A price-comparison web app plus a Manifest V3 Chrome extension, ~95KB with zero dependencies and zero external requests. Your data stays in your browser's localStorage. An exercise in building useful tools with nothing to leak.
Commissions
I design and ship privacy-preserving tools and AI governance controls for Canadian organizations, from heuristic prototypes to production systems. Start with a discovery call.
Ask movahedi.ca
Answers from Mohammad's articles
