Local-first AI tools that run on your own GPU, offline-first ERP and POS systems running in real restaurants and factories, and fixes that ship in open-source projects used by millions.
Each one solves a problem I actually had, or one a client was paying for.
Watches the web pages you care about, works out what genuinely changed, remembers it, and drafts a message, which never leaves your machine until you approve it.
A local coding agent where your compiler, not the model, decides what ships. Several models propose edits; each is compiled in a shadow workspace and your own toolchain picks the winner.
.cavemanrules are enforced by the scorer, not just the promptBilling, KOTs, inventory and reports for an independent restaurant, built so the counter never waits on the internet.
ERP for a build-to-order pharmaceutical machine manufacturer, with Indian GST, written from scratch.
Turns a one-line bug description into a full bug report, and diffs a live page against its Figma design, element by element.
Paste your whole career once and get a resume tailored to each job. Five LangGraph steps, wrapped in code that checks the model's work.
Smaller tools and apps.
Merged contributions to projects other people depend on.
frac in PythonCodePrintermerged*args / **kwargs docstring descriptions on tool parametersmerged--version in the man page and fix --no- option spellingmergedThe tools I reach for, grouped by job.
The same few rules show up in every project.