MindCoach Project Story Inspiration As a lifelong learner, I saw how hard it is to find tailored educational content. Inspired by AI’s potential, I built MindCoach using OpenAI’s gpt-oss-20b to create personalized learning paths for anyone, anywhere. What it does MindCoach lets users input a course title and description—like “Python Basics”—and a quick survey of their knowledge. The app, powered by gpt-oss-20b via Hugging Face’s API, generates five customized lessons with titles, overviews, and objectives, tailored to the user’s needs. How we built it MindCoach is simple yet powerful:
Users submit a course title, description, and survey via a React frontend. Inputs are formatted into a prompt for gpt-oss-20b. The prompt is sent to Hugging Face’s Inference API. The AI generates a structured learning path with five lessons. Lessons are saved as markdown and displayed to users.
Challenges we ran into
Hardware limits: Lacking GPU power, I leaned on Hugging Face’s API, optimizing for cost. Prompt tuning: Early lesson plans were generic, needing refined prompts for specificity. User customization: Balancing survey inputs with concise outputs took iteration. Cohesion: Ensuring lessons formed a logical learning path required prompt engineering.
Accomplishments we’re proud of
Built an intuitive tool for learners of all levels. Leveraged gpt-oss-20b to create precise, tailored educational content. Kept costs low using Hugging Face’s API. Delivered a flexible system for diverse learning goals.
What we learned
Prompt engineering is key for tailored AI outputs. Structured inputs improve response accuracy. API-based workflows bypass hardware constraints. Iterative testing enhances AI performance.
What’s next for MindCoach
Add offline mode for local gpt-oss-20b inference. Integrate video-based lessons via generation APIs. Build a mobile app for broader access. Enable user feedback for iterative lesson refinement.
Log in or sign up for Devpost to join the conversation.