ML Tower
District overview
Browser-game model tests, a sports projection model graded against the market, content-generation helpers, and source records for the studio’s AI and automation experiments.
ConceptDevelopment details
- Source code
- Public and private
Built with
Code repositories
Screenshots
OpenCut Roughcut web interface for starting a local rough-cut workflow.
ClawX dashboard view for managing agent work in a desktop UI.
A Blush Book brand board generated by the private Shorts Pipeline tool.
The Game Hub in the archived Gameplan Turbo prototype, before a project is added.
Models tested through working software
The Platformer Slice Benchmark asks models to build a game, then checks what happens when it runs. Scores sit beside playable outputs and failure notes, including cases where the testing code got the result wrong.
A model graded against the market
Prop Commander projected WNBA player props and compared them with sportsbook prices. Its evaluation harness graded 1,010 props across 13 game dates and found the model carried no information the market did not already have, so the model was retired and the result was published with the dataset.
Tools used around the studio
Shorts Pipeline generates brand boards and call-to-action previews for short-form content. The source records also include planning and editing experiments, a ClawX reference fork, and a media-player-style dashboard skin for Hermes Agent.
Current and archived work
The benchmark, Prop Commander, and Shorts Pipeline have their own project pages. Earlier OpenCut, ClawX, and Gameplan experiments are in the Archive, where their pages distinguish completed work from unfinished prototypes and upstream projects.