Skip to content
View GrandMastaShake's full-sized avatar

Block or report GrandMastaShake

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
GrandMastaShake/README.md

Hey, I'm Alex

I'm teaching a small language model from scratch, the way you'd want a mind raised: rules before opinions, lessons in the order people learn them, and every claim tested before I believe it. Along the way I build the tools, papers and kids' books that the work keeps asking for.

I write down what I expect before I run an experiment, replicate across seeds, and publish my retractions next to the claims they correct. Being wrong out loud is part of the fun.


What I'm working on

Ember (private for now, happy to share on request)
A 16-25M parameter language model I'm building and teaching myself. Every architecture experiment is pre-registered and seed-replicated: 35 notebooks and 57 runs so far, with the retractions kept right beside the results.

blast-radius
My paper-track entry for Kaggle's Gemma 4 Developer Agent competition. A ripple walk over a code graph that tries to predict every file a bug fix will touch. On 129 tasks it beat keyword search at top-3 recall (43.7% vs 41.3%) and lost at top-8 (54.4% vs 68.4%). The write-up reports the loss, the ablations, and why it happened.

enaqt-open-quantum-chains
Code and data for my single-author paper on environment-assisted quantum transport in open chains, submitted to arXiv in October 2026. Every number in the paper regenerates from one script.

EmberArmor
A rule gate that checks an AI coding agent's tool calls before they run. 41 rules, linted with Z3, replayed against about 50,000 real agent calls, with its limits written down in the README.

Wild & Whimsy
182 open-ended children's books for ages 3-9, plus a parent app. No villains and open endings: stories that teach curiosity and kindness without preaching. I think of it as alignment work for humans.


Also built

sector-regime-heatmap
An auditable market research system for the 11 GICS sectors. Every call sits next to the evidence against it, and pre-registered triggers fire when its own assumptions break.

game-component-library
81 game-system contracts and 12 validated compounds that AI agents can assemble into games, with validators that reject bad wiring. It grew out of roblox-modular-lib (49 strict-typed Luau modules) and bits-and-baubles-blender-kit (procedural Blender assets to Roblox-ready FBX).


What I work with

Python Luau JavaScript / TypeScript Git
Hugging Face Z3 Blender Roblox Studio Unity


Stay-at-home dad and restaurant server in central Pennsylvania. Self-taught in machine learning, software and physics, and having a great time with all of it. Hugging Face: GrandMasta1024 | Roblox: GrandMastaShake

Pinned Loading

  1. bits-and-baubles-blender-kit bits-and-baubles-blender-kit Public

    Python

  2. Corporeus Corporeus Public

    Python

  3. EmberArmor EmberArmor Public

    Python 1

  4. EmberBench EmberBench Public

    Python

  5. EmberHoneypot EmberHoneypot Public

    Python 1

  6. roblox-modular-lib roblox-modular-lib Public

    Lua