Making an AI system's reasoning visible, editable, and testable.
I built MindAssembly as a solo hackathon project to explore a question that matters beyond one demo: what changes when people can see the structure behind a machine's conclusion?
1inspectable reasoning graph
ROLE
Solo creator · Product, design, and engineering
WHEN
August 2026
WITH
Independent hackathon project
MindAssembly began with my interest in human thought, algorithms, and trustworthy AI. Instead of building another chat surface, I wanted to expose a small reasoning system as something spatial and inspectable.
The prototype represents concepts as weighted nodes and their relationships as edges. Its behavior is deterministic by design: the goal is not to imitate a hidden model, but to give people a controlled environment for understanding how structure and influence shape an outcome.
MindAssembly turns reasoning into a deterministic weighted graph. Users can inspect nodes and connections, alter influence, rerun the system, and compare how the conclusion changes.
03
THE PROCESS01 02 03
How I moved through the work.
Define the mental model
I reduced the experience to concepts, connections, weights, and an outcome so the interface could teach the system without a manual.
Build the graph engine
I implemented a deterministic weighted-graph simulation in TypeScript, keeping every change reproducible and inspectable.
Turn analysis into interaction
I designed editing and rerun flows so users could ask counterfactual questions by changing the system itself rather than writing another prompt.
04
KEY DECISIONS
The choices that shaped the result.
- Use deterministic behavior so identical inputs always produce the same result
- Represent influence visually instead of burying it in a settings panel
- Build an original reasoning interface rather than a conventional chatbot
- Keep the prototype honest about what it simulates and what it does not
WHAT IT DELIVERED
Built and shipped a complete solo prototype
Made model-like reasoning legible as an interactive system
Designed for counterfactual exploration rather than passive consumption
MY SCOPE
Product concept
Interaction design
Graph simulation
Frontend engineering
MindAssembly clarified the direction I want to pursue: intelligence tools should not only generate results. They should help people inspect assumptions, understand relationships, and reason more deliberately.