Dump-ster Diver

Dump-ster Diver

A Knowledge Graph Forensics System for Document Analysis

Dump-ster Diver is an AI-powered document analysis system that extracts entities, relationships, and insights from large document collections. This tool transforms unstructured documents into an interconnected knowledge graph, making it easier to discover patterns, connections, and key information across thousands of files - whether you're analyzing legal documents, corporate archives, email dumps, or any other large-scale document repository.

View on Github

Features

🎯 Processing Mode

Simple Processing Mode (Recommended for Large Collections)

• Generates concise 1-2 sentence summaries for each document

• Automatic document type classification (email, memo, legal-document, chat, etc.)

• Intelligent tag extraction for easy filtering and search

• Faster processing ideal for initial document review

• Review status tracking and flagging system

📄 Supported Document Types

• Text Files: .txt

• Images: .jpg, .jpeg, .png, .gif, .tif (OCR via vision models)

🎨 Interactive Web Interface

• Windows 95-inspired retro UI for nostalgia and clarity

• Filter documents by type, tags, review status, and flags

• Document detail viewer with inline text/image display

• Real-time processing progress monitoring

• Dark/light mode toggle

🤖 Configurable AI Models

• Note: these are just what I used while testing this project

• Text Models: llama3.2:3b, qwen3:4b-instruct, qwen3:8b, qwen3:30b

• Vision Models: gemma3:4b, gemma3:12b, qwen3-vl:8b, qwen3-vl:32b

• Switch models on-the-fly through the UI

• Powered by Ollama for local, privacy-focused AI

💾 Graph Database Storage

• Neo4j 5.13 with APOC plugins for advanced graph operations

• Efficient querying of entity relationships

• Built-in similarity relationships between documents

• Persistent storage with Docker volumes

---