DoubtLens

About DoubtLens

DoubtLens is a personalized study companion for Track D. A student photographs a textbook page or lecture slide. The app reads that page, teaches the next step, and gives practice from the same source.

Inspiration

Students already have the notes. They get stuck on one move: what this slide is asking, which line of the textbook matters, or what to try next. A lot of AI tools ignore the page and jump to a final answer. We wanted a tutor that stays on the student's own material and helps them keep going.

What it does

1. Upload a photo of a textbook page, lecture slide, or worksheet.
2. DoubtLens reads the visible question and notes.
3. The student picks Hint only, Next step, or Full solution.
4. The app returns:
   • the question taken from the page
   • a short learning goal
   • a hint tied to that page
   • the next useful step
   • a common mistake to avoid
   • two practice questions from the same topic
5. The student can type "I tried this" and get feedback on that attempt.

Full solutions stay hidden until the student asks for them. The hint names the uploaded page as the source, so the help is cited instead of coming from a random web answer.

How we built it

• Frontend: HTML, CSS, and JavaScript, no build step
• Backend: FastAPI with a fixed tutor JSON format
• Vision: Gemini or OpenAI when an API key is present
• Demo mode: bundled textbook-style pages so the product still runs without a key
• Prompting: the photo is the source of truth, and the model should not invent missing work

Challenges

• One photo can contain more than one question
• Glare and handwriting make reading harder
• Models want to finish the whole problem; the product has to stop them
• A useful tutor has to say where the hint came from, not only what the answer is

What we learned

Personalized tutoring is a product problem as much as a model problem. The best help is often one step shorter. Seeing the student's own page matters. Citing that page matters too.

What's next

• Let the student circle one question on a crowded page
• Add a short source line on every hint, such as "from the uploaded slide"
• Turn lecture-slide photos into a small cited study set
• Let practice questions get harder after a correct try

Built With

• computer-vision
• css
• education
• fastapi
• gemini
• html
• javascript
• multimodal
• openai
• pydantic
• python
• uvicorn