Common questions about how this research tool works
It is a free, AI-powered tool that lets you search through publicly released U.S. Department of Justice documents related to the Jeffrey Epstein case using natural language questions. Instead of manually browsing thousands of PDFs, you can ask questions and get sourced answers with direct links to the original documents.
The documents are sourced from the Nikity/Epstein-Files dataset on HuggingFace, which aggregates publicly released records from the U.S. DOJ, the Southern District of New York, and the U.S. House Oversight Committee. These include court filings, depositions, emails, flight logs, and other records entered as exhibits in legal proceedings.
Responses are generated by an AI model based on retrieved document passages. While the system is designed to provide sourced answers, it may occasionally misinterpret documents, omit relevant context, or produce inaccuracies. Every response includes document citations (EFTA IDs) that link to the original PDFs so you can verify the information yourself.
EFTA followed by 8 digits (e.g., EFTA00009654) is the official document identifier assigned by the U.S. Department of Justice. Each ID corresponds to a specific document in the public release. When the assistant cites an EFTA ID, it becomes a clickable link to the original PDF on justice.gov.
No. Your chat history is stored only in your browser's local storage and never leaves your device. When you send a question, only that single message is sent to our server to generate a response — we do not store, log, or retain your queries after the response is delivered. You can clear your local chat history at any time using the trash icon.
The tool uses Retrieval-Augmented Generation (RAG). Your question is converted into a numerical representation (embedding) using OpenAI's model, then compared against 2.2 million document vectors in a Pinecone database. The most relevant passages are reranked using Cohere and passed to a Llama 3.3 70B language model (via Groq) to generate a sourced answer.
You can use this tool as a starting point for research, but you should always verify information against the original source documents before citing it. AI-generated responses may contain errors. This tool is intended for educational and research purposes only and does not constitute legal advice.
The AI model generates responses based on retrieved document passages. In some cases, the model may synthesize information from general knowledge rather than specific documents, particularly for broad or contextual questions. For the most document-specific answers, try asking detailed questions about specific events, people, or dates mentioned in the court records.
No. This is an independent, open-source project with no affiliation to the U.S. Department of Justice, any court, or any government agency. It simply provides a search interface over publicly released documents that are freely available on justice.gov.
The project is open source and hosted on GitHub. You can review the source code, report bugs, suggest features, or contribute at github.com/JuanPagola/epstein-rag-chatbot.