prsist
One memory that Claude, Claude Code and Codex share. It turns your sessions into short notes, files them by project, and gives the right ones to the next chat. Transcripts are deleted within a day.
Senior Data Analyst at SquareTrade Europe. The AI tools I build are measured, tested and cheap to run.
One memory that Claude, Claude Code and Codex share. It turns your sessions into short notes, files them by project, and gives the right ones to the next chat. Transcripts are deleted within a day.
A running coach that reads the runs you already record and adjusts your plan after each one. A rules engine works out the safe options. Claude only picks between them and explains why, so the plan never depends on the model being right.
Finds X posts worth replying to and drafts a reply in my voice. Free rules filter spam first, Claude Haiku scores the rest for about $0.0002 a post, and code checks every draft before I see it. I post every reply myself.
My agent skills for building and fixing code. Each change goes through a break loop where a Claude and a Codex reviewer try to break it, until a fresh round finds nothing.
Case study · GitHubI built the internal tool behind our monthly insurer reports at work. Now I’m rebuilding it as modular, tested code with coding agents, to make it more robust and more efficient.
Work project, no public codeHow my Debian box, MacBook, Mac mini and iPhone run coding agents together over Tailscale, written up so the next machine sets itself up.
Private notesA macOS menu bar app that tracks goals against live Stripe revenue. Planned with test-first task briefs for agents.
GitHubA finance tracker, a habit dashboard and an AI note-taking app. Early full-stack builds.
Ledgerly, Streakly, NoteMindSeven small apps in two months, including a prompt engineering micro-SaaS and a customer feedback analyser. Where I learned the API basics.
Private reposI lead the Finance Analytics function at SquareTrade Europe, which has recently become an insurer. I build the analytics engineering behind our insurance and finance reporting: traceable dbt models, tested Python pipelines and data quality checks. At work I use GitHub Copilot to build AI-assisted validation and automation workflows.
Scoring prsist's note-taking against sessions I labelled by hand, with Claude Opus as the judge, separate dev and test splits, and three repeats per run so I can see the noise.
In cstack, Claude and Codex try to break each change, and a model from the other family settles disputes.
In KM Zero, a deterministic engine builds every option and Claude only chooses and explains. If the model fails, the runner still gets a correct plan.
Tailscale, a shared account hub and T3 Code, so I can start work from my phone and pick it up on my laptop.