DocParse AI Pro

An AI-powered document intelligence assistant that transforms messy lecture PDFs, scans, and textbook photos into structured executive summaries, flashcards, and practice quizzes.

Comment

Inspiration

As students, we are constantly buried under an avalanche of dense lecture PDFs, handwritten slides, and phone photos of textbook pages right before exams. Standard text-parsing tools frequently fail when encountering scanned pages, multi-column layouts, or image-heavy documents.

We built DocParse AI Pro to create a unified, multimodal study engine that transforms any document or image—whether a digital PDF or a quick smartphone snapshot—into structured, actionable study materials in seconds.

What it does

DocParse AI Pro allows users to upload text-heavy PDFs, scanned documents, and photo uploads (.pdf, .png, .jpg, .jpeg). Powered by Google's Gemini 3.8 Flash model, it converts complex files into:

Structured Executive Summaries & Study Notes: Rendered in a clean, printable report preview layout instead of unformatted raw text.

Exam Flashcards & Q&A: Formatted into quick-study pairs for active recall.

Multiple-Choice Practice Quizzes: Designed for immediate revision and testing.

Key features include 1-click Markdown copying, print-to-PDF support (window.print()), and persistent client-side API key management via browser localStorage.

How we built it

DocParse AI Pro was built to be fast, lightweight, and serverless:

Frontend & UI: Built with Next.js, React, and Tailwind CSS using v0.dev for rapid prototyping and executive report preview styling.

AI Engine: Integrated directly with Google's Gemini 3.8 Flash (gemini-3.8-flash) API for multimodal intelligence.

Client-Side File Processing: Uses JavaScript's FileReader API to read uploaded PDFs, PNGs, and JPGs directly in the browser as Base64 streams.

Payload Sanitation: Strips the Data URL prefix (e.g., data:application/pdf;base64,) dynamically before constructing the API payload.

Multimodal API Request: Sends the clean Base64 string and user prompt into Gemini's inlineData REST payload schema:{ "contents": [{ "parts": [ { "inlineData": { "mimeType": "application/pdf", "data": "CLEAN_BASE64_STRING" } }, { "text": "Generate executive summary and study notes." } ] }] } Deployment: Hosted on Vercel with source code synchronized via a public GitHub repository.

Challenges we ran into

Base64 Payload Sanitation: Early API requests failed due to standard FileReader outputs appending headers like data:application/pdf;base64,. Stripping the header prefix dynamically was essential for raw multimodal parsing.

Model Deprecation & Versioning: Adapting to updated API model identifier requirements (transitioning to gemini-3.8-flash) while keeping client-side state reactive.

Formatted Document Rendering: Bypassing plain text outputs by building an executive A4-styled report container complete with copy-to-markdown and instant print-to-PDF functionality.

Accomplishments that we're proud of

Rapid Prototyping & Execution: Building and deploying a fully functional, production-ready web application under strict hackathon deadlines.

Executive PDF Report View: Successfully rendering AI outputs into a clean, printable document template rather than raw, unformatted text.

Zero-Server Overhead: Managing API interactions securely via client-side requests and persistent local user key storage.

What we learned

Client-Side Multimodal Execution: Efficiently handling heavy file data streams purely in the browser without intermediate backend storage.

Secure Local Key Management: Utilizing browser localStorage for API key persistence, offering zero-server friction and user privacy.

AI-Assisted Workflow: Leveraging generative UI tools to rapidly construct and iterate complex full-stack web applications.

What's next for DocParse AI Pro

Multi-Document Batch Analysis: Cross-referencing multiple PDFs and lecture decks simultaneously.

Spaced-Repetition Tracking: Implementing active recall schedules for generated flashcards.

Ecosystem Exports: Direct 1-click export support for Anki, Notion, and Quizlet.

Built With

Share this project:

Updates

Submission history