Epstein's Phone (Reconstructed) is an innovative Progressive Web App (PWA) designed to seamlessly browse emails contained in the Jeffrey Epstein estate documents released by the U.S. House Committee on Oversight and Government Reform. This project converts a large volume of email screenshot images into a well-structured, searchable database while providing an intuitive interface reminiscent of iPhone Messages.

Features Include:

• A live demo accessible at epsteinsphone.org
• Full mobile optimization for an enhanced user experience on smartphones
• Offline functionality to browse data without internet connectivity

Project Status

• Fully Implemented Features:

  • Utilizes a Vision LLM-based extraction leveraging Qwen 2.5 VL 72B to process email screenshots, resulting in structured outputs.
  • Built on an optimized SQLite database, housing 3,997 unique messages, complete with sender/recipient information, timestamps, subjects, and HTML content.
  • Implements a parallel processing approach for swift data handling, employing a 20-thread system to enhance performance.

• Progressive Web App Characteristics:

  • Simulates an iPhone device frame, complete with a realistic status bar.
  • Offers iOS-style navigation with sections for Messages, Photos, and Documents for a familiar user interface.
  • Fully installable on mobile devices, allowing users to add the app to their home screens.
  • Utilizes a service worker to provide full offline capabilities post initial loading.

Key Achievements

• A client-side email browser that operates entirely in the browser via WebAssembly without a backend server, mimicking the authentic iPhone experience.

Dataset Overview

Based on 23,124 JPEG images extracted from Epstein estate documents, this project has established a comprehensive message database with:

• Structured message timelines reflecting conversation flow.
• Extensive searchable content across all messages.
• Thread reconstruction enabling easy navigation through related messages.

Technical Architecture

The robust data processing pipeline integrates the following stages:

1. Email Identification: Filtering relevant text files.
2. Image Processing: Efficient resizing and encoding for processing.
3. Vision LLM Extraction: Extracting structured email data from images.
4. Data Cleaning: Standardizing names, removing duplicates, and enhancing overall quality.
5. Database Storage: Facilitating efficient querying through a well-structured schema.

Technology Stack

• Python 3.x utilizing OpenAI for backend processing.
• SQLite for local data storage.
• JavaScript and CSS3 for frontend development, ensuring optimal performance and user experience.
• SQL.js for client-side SQLite database management.

Future Enhancements

Plans for additional features include refining conversation threading logic, enhancing profile pictures for contacts, and further optimizing the database for superior performance.

Documentation & Resources

This project is available for further exploration and educational purposes, with codebase access on GitHub. All data used is publicly sourced from government documents, ensuring transparency and legal compliance.

The ready-to-deploy structure with no dependencies makes it easy for anyone to contribute or utilize in their projects. Engage with this essential resource for those interested in exploring discussions around prominent figures, while encouraging ethical research practices.