I build tools that turn scattered, implicit knowledge into something you can read, hand to someone else, and still trust three weeks later.

That sounds abstract, so here is what it actually means in three cases.

The tags on your tests already describe what you want to run. Nothing turns that into a decision you can defend.

Compose a precise selection from your tag taxonomy, freeze it into a portable run spec — a versioned document holding the rule you wrote, the exact list of tests it resolved to, and a fingerprint of the suite at that moment — and let anything execute it: a local process pool, your own agent on a remote machine, a CI job. Results come back from any transport into a history that feeds the next selection.

The interface is a document, not a protocol. So the next morning, when someone asks what you actually ran, the answer is a file rather than a memory.

Python · MIT · web UI, API and CLI

Your test files already explain themselves. Nothing was rendering it.

A VS Code extension for Robot Framework that turns the explanations sitting in your .robot files — [Documentation] blocks and ordinary comments — into rendered prose in a side panel that follows your cursor, where every line clicks back to the exact source line it came from. And because a test is only half prose, it reads your Python keyword libraries too, and tells you what a keyword actually returns, as a path you can paste.

It adds no formatter, no diagnostics, no language server. It sits beside whatever tooling you already run, and it reads your Python without ever importing or executing it.

JavaScript · MIT · v1.0.0 on the Marketplace, in daily team use

Agent memory doesn't decay, it accumulates. Duplicates drift, facts go stale, and the index bloats until the rules drown in their own context.

A tiered memory engine for file-based agents, built on how human memory actually works rather than as a metaphor: a capture inbox as scratchpad, curated notes as episodic store, guidelines and core rules as semantic store, and a grep-able archive where nothing is ever deleted. Offline consolidation does the work sleep does — replay, dedupe, promote, decay, re-index — so the bank has a metabolism instead of being a landfill.

Python · alpha, in daily production use

Each one takes something that already exists but isn't legible — tags scattered through a suite, explanations buried in comments, knowledge spread across sessions that end — and makes it explicit, inspectable, and durable enough to hand over.

They also share a bias: claims carry their evidence. Every capability in those docs is demonstrated by something runnable, not asserted in a feature list. Where a tool's behaviour is narrower than its name suggests, the documentation says so rather than letting you discover it.

Currently self-employed and open to work — test infrastructure, developer tooling, and the unglamorous parts of making an engineering team's knowledge survive contact with time.

Reach me through issues on any of these repositories.