Ben Meyer
Research and software projects, designed and directed by me and built with AI coding agents, mostly Claude Code
Each project started as a question or a need of mine. I did the research, set the design and the rules, and directed an AI coding agent, Claude Code, through the build. Every page says what the thing does and the thinking behind it.
benmeyer753@gmail.comLinkedIn
Projects
- Quant Equity Research PlatformData app · Sep 2026
A research platform built to test whether a stock's technical state predicts its return tomorrow, with rules fixed before each experiment and a live daily record as the only evidence. This is its research browser, exported live, plus a SQL console in your browser.
- FitLogiOS app · Sep 2026
An iPhone strength-training app I designed around a coach that plans the week with you, with rules I wrote and tested on my own training. The coach suggests; the app enforces.
- Tucson Grid & Battery ModelData app · Sep 2026
My study of what battery storage does to Tucson's grid peak, from a year of public hourly demand data. Explore the year and size a battery yourself.
- Company Research AgentData app · Sep 2026
A due-diligence method I designed to stop research from anchoring on early numbers, packaged as a repeatable AI workflow. Run end to end on a real California startup, with the dossier, the evidence registry and the model published in full.
- SOC Volatility DashboardDashboard · Dec 2025
My research into whether the physics of avalanches describes intraday volatility, built into a dashboard that scores stocks on it and tests the signals. The page runs the model and carries a real scoring run, with the validation results as they came out.
How I work
Every project here started as a question I actually wanted answered: whether battery storage pencils out for Tucson’s grid, whether the physics behind avalanches says anything useful about a stock, whether I could get a training plan that would hold me to it. I do the research first, and I write the rules down before there is any code to bend them.
The part I care most about is deciding what would count as evidence before I see the results, and saying out loud what would prove me wrong. On the research platform every experiment is registered before it runs and the model is frozen, and the site reports that it has only 73 of the roughly 250 live days it needs — a number that isn’t earned yet shouldn’t be shown as if it were. On the volatility work I checked the signal against the volatility that actually followed, then against real option quotes with trading costs, because a chart that looks good in hindsight isn’t a result.
Then I use the thing I built and let that change it. FitLog first planned every training day from scratch; a fixed weekly program turned out to serve me better, so I had the daily planner cut rather than kept.
Claude Code wrote most of the code here, and I’d rather say that plainly than let it be inferred. What’s mine is the question, the research behind it, the rules and the specification, the decisions the build kept raising, and the review of every result, including the ones that didn’t say what I wanted. The direction and the judgment are mine; most of the typing isn’t.