Problem
Game design documents often contain mechanics, story, UI notes, worldbuilding, systems, and production concerns in the same long document. A developer or small team can lose time manually pulling those details into actionable categories.
Case study
From design document to structured production map.
AYO Chat Agent v1 demonstrates how local AI can support game design planning without turning the whole workflow into an uncontrolled prompt experiment.
Solution
AYO loads design text, splits it into manageable chunks, sends controlled prompts to a local AI model, checks responses for obvious failure conditions, assigns useful categories, and writes a production map.
Local Proof
AYO is the packaged desktop demonstration of the workflow. It proves the loader, chunker, verifier, categorizer, and output assembly in a reviewer-testable local environment.
Hosted Proof
Matteo is the hosted companion example. It shows the same constrained-model mindset expressed as a live browser product with a clear behavioral contract.
System Components
Loader
Reads input from a selected text file or command-line argument and normalizes the text for processing.
Chunker
Splits long design content into smaller sections so local model calls stay manageable.
AI Provider
Routes prompt text into a local Ollama and Phi-3 workflow.
Verifier
Rejects blank, too-short, and error-like generated responses before they are accepted.
Categorizer
Uses Strategy Pattern logic to assign chunks into production categories.
Production Map
Assembles categorized content into a structured Markdown planning document.
Technical Stack
- Python desktop app with shared pipeline logic
- Ollama and Phi-3 local model workflow
- pytest test suite
- Windows EXE demo package
- UML architecture documentation
- Markdown production-map output
Verification
The test suite covers loader behavior, chunking behavior, response verification, categorization, production-map assembly, and integration-level workflow behavior.
Current verification: 28 tests passing.
Why It Matters
This project demonstrates more than prompt usage. It shows a controlled technical workflow where local AI, software architecture, testing, and game production thinking work together. That supports technical design, AI tooling, developer tools, production systems, and teaching-oriented roles.