JC Davis

Austin, Texas

JC Davis


I build AI products — embedding systems, forecasting models, and the games and tools that sit on top of them. Real data, full scale, shipped.

The work

Flagship work

Built and shipped

Three things I've built and shipped, in sequence.

  1. eye.jcamd.com

    Embedding Atlas

    One 3D map, many layers — jobs, places, water, climate, companies, client datasets — plus a constructs world: every jurisdiction, district, watershed and ecoregion over a point, joined back to the physical layers by location.

  2. jcamd.com/dottie

    Dottie

    An orchestration platform built around a measured loop: one router sends each goal to the cheapest tier that can do the work, and System One answers typed questions as Choice, Score and Noul, served locally by dottie-os.

  3. Dottie · apps/scout-cli

    scout

    Dottie's command line. The harness gets one tool, and every capability is a plugin behind it, each declaring what it may touch. Route a goal, run it, and read the measured timeline.

Timeline

Shipped

2026

  1. 01

    Vector Hoops

    Embedding-map basketball strategy game — 12,966 player-seasons, MTNN v5 champion model, with daily games, pack battles, and a twenty-questions mode.

  2. 02

    Embedding Atlas

    One 3D map, many layers — aircraft, ships, satellites, earthquakes, launches — plus jobs & wages across 393 US metros on real labor data.

  3. 03

    Dottie

    Orchestration platform with a measured loop: one router, the System One decision model, and the local dottie-os sidecar.

  4. 04

    Vector game suite

    The same embedding-map engine across gridiron, pitch, equities, and arcade — one geometry, many games.

  5. 05

    scout

    Dottie's command line: capability-declared plugins behind one entry point, and the harness's front door.

Craft

How I work

  1. I

    Real data, full scale.

    No synthetic demos, no mock results. If it can't run on production data, it doesn't ship.

  2. II

    Held-out evaluation.

    Challengers earn promotion with measured wins on unseen data — never on vibes.

  3. III

    QA at every step.

    Pipelines with gates: a failure stops the line, it doesn't get papered over.

  4. IV

    Products, not decks.

    The proof is something people can touch — a game, a map, a tool.

Lab

Experiments

The vector games

dumbmodel.com
  • hoops.dumbmodel.com

    Vector Hoops

    A daily NBA puzzle played on a live 3D embedding map of 12,966 player-seasons: daily guessing games, pack battles, twenty-questions mode, and client-side inference.

    Play Vector Hoops
  • gridiron.dumbmodel.com

    Vector Gridiron

    A fantasy football cockpit that ranks weekly lineups with next-game predictions from a multi-tower net trained on nflverse data.

    Play Vector Gridiron
  • pitch.dumbmodel.com

    Vector Pitch

    A daily World Cup puzzle over StatsBomb Open Data: 633 player-tournaments from the 2018 and 2022 World Cups.

    Play Vector Pitch
  • equities.dumbmodel.com

    Vector Equities

    An interactive embedding map of public companies built from SEC EDGAR XBRL filings, with a sector-coherence eval.

    Explore Vector Equities
  • vector-unified.vercel.app

    Vector Unified

    Player embeddings across leagues on a single sphere: the engine behind the vector games.

    Explore Vector Unified

Tools

  • jcamd.com/graphify

    Knowledge graph viewer

    A static, explorable viewer for knowledge graphs — browse entities and relationships in the browser.

    Open knowledge graph viewer
  • jcamd.com/chips

    CPU · GPU · DPU, exploded

    Three processors exploded in interactive 3D: EPYC 9654, H100, BlueField-3. Every block labeled, every difference explained — ELI5 to expert.

    Open CPU, GPU and DPU exploded view

Uses

The stack

Python stdlib-first
pipelines, models, tooling
TypeScript + Next.js
product surfaces
Vercel + GitHub Actions
shipping
PyTorch
GPU training box
SQLite
local data
TimesFM
forecasting

Contact

The best way to reach me is GitHub.

Open an issue or discussion on any repository — I read everything. For advisory work, see Alamo St Advisors.

github.com/jcdavis131