A simple Python framework that orchestrates a multi-step research workflow using a state graph, with built-in human-in-the-loop interruptions and automated report generation.
It's a lightweight, multi-agent system around chat models that customizes the research process.
Source Selection
- Users can choose any set of input sources for their research.
Planning
- Users provide a topic, and the system generates a team of AI analysts, each focusing on one sub-topic.
Human-in-the-loopwill be used to refine these sub-topics before research begins.
LLM Utilization
- Each analyst will conduct in-depth interviews with an expert AI using the selected sources.
- The interview will be a multi-turn conversation to extract detailed insights as shown in the STORM paper.
- These interviews will be captured in a using
sub-graphswith their internal state.
Research Process
- Experts will gather information to answer analyst questions in
parallel. - And all interviews will be conducted simultaneously through
map-reduce.
Output Format
- The gathered insights from each interview will be synthesized into a final report.
- We'll use customizable prompts for the report, allowing for a flexible output format.
