The product and the model.

I am a full stack engineer at Codemancers, embedded with Grit Financial, a US fintech, and working on Codemancers' own products alongside it.

Grit lets people draw pay they have already earned, and issues the cards it lands on. It reaches those people through whichever payroll provider their employer already uses, so most of my work there is the layer that lets one platform present itself as several: keeping employee and employment data in sync across providers, the rules deciding who can take an advance and how it gets repaid, and the path someone actually moves through to sign up, from identity verification to a bulk enrollment pipeline that turns an employer's spreadsheet into verified accounts with cards on the way. I also build the portals their employers and operations team work in.

At Codemancers I work on Gather, our standup and meeting product. I built the bot that joins a video call, records it, and transcribes who said what as it happens. Most of the work there was staying in the call rather than getting into it. On top of that sits an agent layer on LangGraph that turns a recording into summaries and action items, and answers questions about the meeting from inside Slack.

I write about the same things: retrieval, LangGraph, and how to test an agent before it goes anywhere near production. What I care about is unglamorous: how a system fails, what it costs to run, and whether the person using it can tell what just happened.