Thinketh

HackGT 13, Georgia Tech · Sep 2026

A personal learning system that teaches you only what you don’t already know.

Stefen Ewers and I built Thinketh in about 36 hours at HackGT 13, in the AI/ML track. It compares what is changing in a field with what you have shown you understand, then teaches the difference.

The idea

Reading something isn’t the same as understanding it. Thinketh keeps two states side by side: what is changing in the world, and what you have demonstrated. A model helps phrase and explain, but it never decides what you know. Mastery moves only through a fixed rule, and every change keeps its evidence and its reason.

The product

Thinketh’s Today screen: 6 new things worth knowing, with counts of items scanned, new, connected to your Mind, and filtered, and a Catch me up button.
Today. What changed in the fields you follow, measured against your Mind.
A Thinketh development on persistent agent memory: the change in a minute, the catch, what it builds on that you knew, why it matters to you, and a Check my understanding button.
A development. The change, the catch, and what it builds on that you already knew.
Thinketh’s Check my understanding screen: a scenario question with four options, one selected, and feedback explaining why it is right.
The check. One question that tests the idea, not the wording.

Demo

What I built

  • The first version of the intelligence layer: the knowledge-state engine and its update rule, adaptive diagnostic selection, the delta engine, and the daily brief.
  • The Hono API in front of it, serving shared Zod contracts to the app.
  • Adapters for every outside service, each with a timeout and a deterministic fallback, and taking MongoDB Atlas with Atlas Search, Tiger Data, Backboard, and Supabase live.
  • Visualize this: a topic-aware diagram planner and a native renderer that draws a change as a diagram.
  • Deploy readiness and demo polish, including a Deno-verified Edge Function, secrets scripts, and fixes from a full click-through of the demo.

Stefen built most of the mobile app, the Playground where two people’s agents teach each other, and the Mindprint layout engine.

Thinketh’s Recently changed list for the Mind: entries labeled Stronger evidence, Unchanged, More certain and Weaker, each with the engine’s reason and when it happened.
The engine at work. Every change to the Mind carries its own reason, even when nothing changed.
Thinketh’s Visualize this screen titled From vanishing context to a curated store, contrasting context as memory, where knowledge is discarded when a session ends, with memory as a store, where distilled facts are consolidated.
Visualize this. A change drawn as the old model and the new one.
Team
Stefen Ewers and Nadani Dixon
Window
About 36 hours
Stack
TypeScript, React Native, Hono, Zod
Status
Working hackathon prototype

Source on GitHubStefen’s engineering case study