This repository provides a synthetic data generation pipeline, context-1-data-gen, designed to create multi-hop search tasks across diverse domains. It targets researchers and engineers who need realistic, multi-step retrieval datasets for evaluating complex information-seeking behaviors. The pipeline generates data following an "explore → verify → extend" pattern, offering a structured approach to simulating intricate search scenarios.

How It Works

The core methodology involves generating multi-hop search tasks by applying an "explore → verify → extend" pattern within each supported domain. This structured approach allows for the creation of complex, multi-step retrieval challenges that mimic real-world information discovery processes, enhancing the evaluation of search and retrieval systems.

Quick Start & Requirements

• Installation: Use uv sync for base dependencies, or uv sync --all-extras to include optional features like reranking, patents, and indexing.
• Configuration: Copy .env.example to .env and populate it with necessary API keys.
• Prerequisites: Requires API keys for Anthropic, Serper, Jina, OpenAI, Chroma, and Baseten, which are used across different domains and functionalities (e.g., embeddings, indexing, reranking).
• Documentation: Detailed documentation for each domain (Web, SEC, Patents, Email) is available in their respective README files.

Highlighted Details

• Supports data generation for multiple distinct domains: Web, SEC filings, Patents, and Email (Epstein).
• Generates synthetic multi-hop search tasks following a structured "explore → verify → extend" pattern.
• Context-1 model weights are available separately.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord/Slack), or project roadmap are present in the provided README snippet.

Licensing & Compatibility

The license type and any compatibility notes for commercial use are not specified in the provided README snippet.

Limitations & Caveats

Operation is heavily dependent on numerous third-party API keys, potentially leading to significant costs and external service dependencies. The project's description is listed as "None," suggesting it may be in an early or incomplete state.