Index

Code

These are the working surfaces behind the products, grouped by the question each one was built to answer. Most are private, because the products are on the App Store and the trading system holds live keys, so a private row publishes its language and its dates and nothing else. Every detail page carries one piece of its actual code, chosen for the decision it shows. Nothing here is padding: duplicates, scratch experiments and dead ends were removed by name rather than listed.

19 repositories · 3 languages · 5 threads

19

repositories

8,524

commits

169

days with a commit

3

languages

Measured 26 August 2026

Filter

Showing all 19 repositories

Market systems5

This thread starts in May 2022, with paid work: SQL over consumer purchasing data at Sports Innovation Lab, fan personas across four leagues, and a report on casino spend. Four years of the same question asked better each time, what is this actually worth and am I right, from that to prop-line analysis to arbitrage scanners to a live quant system that trades its own money. The money keeps score whether or not I agree with it.

  • arborPRIVATE    A quant trading system on Kalshi that trades its own money.Python · 7 days agoRead
  • singal-flowSignalFlow — the equities predecessor to arbor: six agents, Kelly-criterion sizing, and supervised learning from its own trade outcomes.Python · 3 months agoGitHub (opens in a new tab)
  • polymarket-kalshi-btc-arbitrage-botPRIVATEThe same idea narrowed to the Bitcoin hourly market, where the two venues list genuinely comparable contracts.Python · 5 months agoPrivate, and with no detail page — nothing to link to
  • poly-kalshi-arbPRIVATEThe first arbitrage scanner: read two prediction markets, find the same event priced differently, report the spread. Never placed an order.Python · 8 months agoPrivate, and with no detail page — nothing to link to
  • SportsBetPRIVATE    NFL prop and parlay analysis with a natural-language query layer over computed statistics.Python · 9 months agoRead

Consumer AI6

Products where a model does something specific and checkable, rather than being the feature. The shipped ones are on the apps page; these are the rest of the attempts.

  • movieWatcherPRIVATE    KinoLog — a film diary that commits to a rating before you watch, then measures itself against what you actually thought.TypeScript · 3 days agoRead
  • recipe-fixPRIVATE    RecipeFix — adapts any recipe to a dietary constraint without wrecking the dish, and exposes the engine as an MCP server.TypeScript · 3 days agoRead
  • intelligent-training-osPRIVATE    Bodivo — adaptive training that models the individual rather than the plan, and says so when it does not know.TypeScript · 3 days agoRead
  • catchtheliePRIVATE    Alibi — a daily logic puzzle, and the calibration pipeline that keeps its difficulty steady.TypeScript · 4 days agoRead
  • ScrapbookPRIVATE    Margin — upload a batch of photos and it builds a narrated chapter with a visual identity generated for it.TypeScript · 3 weeks agoRead
  • queuedPRIVATE    MammaMxir — a Spotify-powered Discord music queue with a web interface, because bot queues have none.TypeScript · 3 months agoRead

Distribution2

Building the thing is half of it. This is the other half — programmatic SEO and affiliate economics, where the engineering exists to answer a question about a market rather than a user.

  • FreelanceStack    Programmatic SEO: thirty tools scored on five weighted dimensions, generating 178 pages including 62 head-to-head comparisons.TypeScript · 4 months agoRead
  • pitchlessaiPRIVATE    Pitchless — scores Reddit posts for whether your product genuinely answers them, and drafts a reply worth sending.TypeScript · 4 months agoRead

Infrastructure2

Software other software depends on. Both of these exist because a guarantee was missing — that a retry cannot post twice, and that a bug which never throws still gets found.

  • Halyard    Social publishing infrastructure built around one rule: a retry must never post the same thing twice.TypeScript · todayRead
  • sentinelPRIVATE    Watches a deployed app for the bugs that do not announce themselves, and proves one is real before reporting it.TypeScript · 6 weeks agoRead

Prototypes4

Single-burst projects: an idea taken far enough to see whether it was worth continuing, and then not continued. Listed because the ratio is the honest part.

Public rows live from GitHub, cached for an hour. Private rows are metadata only.