Primary Developer · 2026
AI-Study-Companion
Built a hackathon learning app and constrained local AI responses to formats the application could use.
- React
- Node.js
- Express
- SQLite
- Ollama
- Learner answer
- Local model
- Output checks
- Feedback
Overview
Our team built AI-Study-Companion for Global Hackathon 2026, whose theme focused on the future of learning with AI.
My role
As the primary developer, I built and operated the application and integrated a locally hosted model through Ollama.
Problem
The application needed predictable model responses to display feedback and evaluate short math answers. Backend timeouts and duplicate definitions also interfered with the API.
Approach
I used harness engineering to constrain model responses to the required formats. I added verification for short math answers and resolved the timeout and duplicate definition issues in the backend.
Outcome
The learning flow could use the structured model output to present feedback and check short math answers. The code and project structure are available in the team repository.
Lessons learned
Define what the application expects from the model and handle failures at the integration boundary.