Epstein Explorer
Document investigation platform with AI entity extraction
The problem
Thousands of pages of declassified material are public but practically unreadable - entities span filings, relationships are implicit, and timelines are fragmented. Journalists and researchers needed an actual investigation tool, not another PDF viewer.
Our approach
We built an ingestion pipeline that extracts entities, relationships, and events from every document, then exposes the graph through search, filtering, and an interactive network visualization. A retrieval-augmented chat layer answers questions with every claim cited back to a source passage.
Public platform with ongoing document ingestion.
Pages ingested, indexed, and cross-referenced automatically.
Semantic retrieval with every AI answer linked back to a source passage.
What shipped
AI entity extraction
LLM-powered extraction of people, orgs, locations, and events with confidence scoring and human review queues.
Network visualization
Force-directed graph of connections, filterable by date, document, or entity type.
Source-cited AI answers
Semantic retrieval over a Bedrock Knowledge Base (S3 Vectors) grounds an AI chat layer - every answer links back to the document passage it came from.
Tech stack
Services behind this work
Further reading
Want something like this?
Book a free 30-minute call. We will give you honest feedback on feasibility and scope.
Start a conversation