Dill.study

Home screen where users can login with Google and upload/link PDFs of their homework
Problem parser for user to verify that the automated PDF parsing correctly extracted questions
Problem dashboard to see each question, click on them for individualized help, and filter by problem status
Main functionality-- chatbot to give step-by-step guidance towards solution, hint buttons, and similar question generation. Tutoring style
Past uploads dashboard with assignment due dates and options to send reminders via AgentMail

Inspiration

Students often lose time and struggle with dense PDF assignments instead of solving problems. Events like office hours may not always be accessible because of time or location barriers. We wanted to remedy this by creating a focused workspace tool.

What it does

Upload a PDF → extract text → split into structured problems. Estimate time per problem. Let users track attempts, hints, status, and assistant interactions. Provide AI problem segmentation & guidance. Tag problems automatically with the problem type and content. Suggest additional problems to enhance learning certain topics. Run reminder + email automation scripts to keep users on pace.

How we built it

Frontend: React + Vite + TypeScript. PDF Parsing: Custom extractor + worker + heuristic parser with AI overlay. AI: Gemini API with fallback models. Auth/Persistence: Firebase. Automation: Scripts for reminders, seeding, drafts, and live send (AgentMail).

Challenges we ran into

Cleaning up messy PDF text (page breaks, numbering noise). Building heuristic parser solid so it works even without AI. How to get realistic time estimates.

Accomplishments that we're proud of

We are proud of the seamless degradation features, the fact our app can work even during Gemini outages. The hints and AI chat effectively help users without ruining the learning experience by revealing too much. The finished product allows for users to easily work on multiple homeworks at once using the data storage features.

What we learned

We learned a lot about effective PDF parsing and OCR in order to intake hard to read PDFs. Pairing this with the use of AI we learned how to get the desired results consistently from somewhat unpredictable models, giving our product clean and reproducible outcome.

What's next for Dill.study - For when you're in a pickle

Next we want to add conversational features with ElevenLabs so users can have natural conversations with the AI. Past that we see ourselves taking many routes, potentially adding features like a friends list and class competitions. All this to say, the future looks bright for Dill.study!