Ai tutor

Inspiration

Many students struggle to convert lecture notes into practical understanding. Most tools either summarize content too briefly or do not support interactive learning after the summary. The inspiration for Ai Tutor came from this gap: we wanted one workflow where a student can upload study material, get a clear explanation, and continue learning through guided follow-up questions.

What it does

Ai Tutor lets users upload a PDF or image file and receive a structured learning response. It explains key ideas, gives real-world relevance, and supports follow-up Q&A in a conversational format. It also generates multiple-choice questions based on the uploaded material to help with quick self-assessment. The system is designed to provide plain, easy-to-read responses and keep the experience simple for students.

How we built it

We built Ai Tutor as a full web application with: A backend API layer for file handling, prompt orchestration, and model communication A custom frontend interface for uploading files, viewing generated lessons, and asking follow-up questions Text and image extraction logic for learning materials Prompt engineering to produce structured educational outputs Runtime safeguards such as model fallback, response sanitization, and error handling for high-demand API scenarios.

Challenges we ran into.

Model availability mismatches across environments and API versions Temporary API overload errors during generation Inconsistent response formatting from model output Handling long responses that got cut off Frontend fetch failures due to transient connectivity/runtime timing issues Balancing richer output (like quizzes) with faster response time

Accomplishments that we're proud of

Successfully migrated from a basic prototype to a backend + frontend architecture Implemented robust fallback logic so the app can recover when a model is unavailable Improved user experience with loading states and cleaner response presentation Added automatic MCQ generation tied to uploaded content Strengthened reliability through retries, health checks, and better runtime controls Delivered a practical learning tool rather than just a one-shot summarizer

What we learned

Reliability is as important as model quality in real AI applications Prompt design must be reinforced with output post-processing API failures are normal in production and must be handled gracefully UX details (loading feedback, clear errors, response readability) strongly affect perceived quality Faster, simpler generation pipelines often outperform heavier multi-call flows for real users

What's next for Ai tutor

User accounts and saved study sessions Topic-wise progress tracking and performance analytics Adaptive quizzes with difficulty levels Support for more document types and multilingual tutoring Streaming responses for faster perceived output Teacher dashboard for classroom usage and assignment generation

Built With

• fastapi
• gemini
• javascript
• python
• tailwind