Inspiration
As engineering students, we often face a common dilemma: missing a crucial lecture due to illness, conflicting schedules, or simply zoning out during a complex explanation. Re-listening to hour-long raw audio recordings is tedious, and staring at static PDF slides without context is confusing. We wanted to build a bridge between the professor's voice and the course materials. We built DeepStudy to act as the ultimate AI classmate—one that never sleeps, captures every detail, and organizes chaos into structured knowledge.
What it does
DeepStudy is a "Neural Academic Engine" that transforms raw course materials into comprehensive study guides.
Multimodal Ingestion: Users upload lecture recordings (Audio) and course materials (PDF slides/textbooks).
Neural Mapping: Using Google Gemini 3 Flash, the system analyzes the "DNA" of the course, breaking it down into logical, digestible modules.
Intelligent Synthesis: It cross-references the audio explanation with the visual data in the PDFs to generate detailed notes, complete with LaTeX mathematical formatting and Markdown structuring.
Cloud Neural Bank: All generated notebooks are synced instantly to Supabase, allowing users to build a persistent library of "Neural Sections" accessible anywhere.
Adaptive Depth: Users can choose between "Standard" processing for quick summaries or "Deep-Scan" for comprehensive mastery.
How we built it
Frontend: We built a futuristic "Neural Terminal" interface using React 19, TypeScript, and Vite. The UI is styled with Tailwind CSS to be dark, sleek, and distraction-free.
The Brain (AI): The core intelligence is powered by the Google GenAI SDK (@google/genai). We utilize the gemini-3-flash-preview model for its massive context window, allowing us to feed raw audio files (converted to Base64) and large PDFs simultaneously.
Orchestration: We implemented a two-step AI pipeline: first, an "Orchestrator" prompt generates a JSON structure of the course topics, followed by a "Synthesizer" loop that generates content for each topic.
Backend: We used Supabase for Authentication and Database storage ("Cloud Neural Bank") to save user progress and generated notebooks.
Challenges we ran into
Rate Limiting & Quotas: managing the API calls to Gemini during the "Deep-Scan" mode (which generates 12+ sections) triggered rate limits. We implemented a robust withRetry logic with exponential backoff to handle 429 errors gracefully.
Multimodal Data Handling: converting browser File objects (audio and PDF) into the specific Base64 format required by the GenAI SDK inline data method was tricky to perfect across different file sizes.
Structured Output: Getting the AI to consistently return valid JSON for the table of contents while returning rich Markdown/LaTeX for the actual content required careful prompt engineering.
Accomplishments that we're proud of
The "Deep-Scan" Engine: Successfully coordinating the AI to autonomously break down a long lecture into 12 distinct, deep-dive chapters without user intervention.
Seamless Cloud Sync: The real-time "Auto-Saving" indicator that syncs AI generation directly to the database as it happens.
Mathematical Rendering: Successfully integrating MathJax to render complex engineering formulas (LaTeX) directly inside the AI-generated notes.
What we learned
We learned how to leverage the Gemini 3 multimodal capabilities to process audio and text in a single context window, which is significantly faster than transcribing audio to text first.
We deepened our understanding of Supabase RLS (Row Level Security) to ensure user data remains private.
We learned how to build resilient UI states (Orchestrating vs. Synthesizing) to keep the user informed during long AI generation tasks.
What's next for DeepStudy : Quickly Understand Missed Lectures
Real-time Live Mode: Processing audio chunks in real-time as the lecture happens.
Quiz Generation: Automatically generating exam questions based on the "Neural Sections."
Multi-language Support: Fully implementing the French/English toggle for our bilingual engineering curriculum.
Built With
- genai
- google-gemini
- react
- supabase
- tailwind-css
- typescript
- vite
Log in or sign up for Devpost to join the conversation.