Secure, AI-Powered Document & Bookmark Vault
I built this project to solve the problem of securely managing sensitive information—from bookmarks to personal documents—without trusting a third-party service. My goal was to create a zero-knowledge tool that uses AI to eliminate tedious data entry and make organization effortless, giving users full control over their data.
Project Overview
Developed a secure, serverless application that transforms a user's Google Drive into an end-to-end encrypted vault. It leverages Gemini AI for natural language management and client-side OCR to automatically extract data from uploaded documents, such as IDs, receipts, and certificates.
Challenges & Solutions
Challenge: Implementing a 'zero-knowledge' security model to ensure absolute user privacy.
Solution: I architected a client-side encryption flow using CryptoJS. All data is encrypted with a user's Master Password *before* being transmitted to the Google Drive API. This means the stored files are unreadable ciphertext to anyone but the user, achieving true data sovereignty.
Challenge: Automating data entry from various uploaded documents like IDs and receipts.
Solution: I created a client-side OCR pipeline using Tesseract.js that pre-processes images in-browser (enhancing contrast and converting to grayscale) for accuracy. The extracted text is then passed to a Gemini AI model with an expert-tuned prompt, which intelligently parses and structures the data into key-value pairs.
Challenge: Ensuring the AI interaction for managing the vault was both powerful and reliable.
Solution: I engineered a sophisticated prompting system for the Gemini API that combines a 'knowledge base' of common document fields with a 'few-shot' example. This guides the AI to produce consistent JSON output, which is then handled by a robust parsing layer that translates the AI's response into executable application commands.
Key Features
- Zero-Knowledge Architecture with client-side AES encryption on Google Drive.
- AI-Powered OCR to automatically parse and structure data from uploaded images.
- Natural Language Management using voice or text commands via Gemini 2.5 Pro.
- Advanced Security including 2-Factor Authentication (2FA) and configurable inactivity timeouts.
- Optimized and Responsive UI built with Next.js and Framer Motion.
- Secure Authentication using Google OAuth 2.0 with minimal, privacy-focused permissions.
Tech Stack
Architecture
I chose Next.js for its performant hybrid of Server and Client components, enabling a fast initial load and a highly interactive UI. To maintain a truly serverless and cost-effective backend, the Google Drive API serves as a personal, encrypted database for each user. This innovative approach offloads the cost and complexity of database management while placing the user in full control of their own data.
Gallery
Explore More Projects
Virtual Assistant for Individuals with Disabilities
AI-powered desktop assistant for blind, deaf, and non-verbal users with speech-to-text, image recognition, and hands-free control.
Employee Management System
The Employee Management System (EMS) is a MERN-stack web app for efficiently managing employees, teams, leaders, and company targets.
Build a Duolingo Clone
Interactive platform for language learning. Features include user registration, course management, quizzes, and progress tracking.