How hkuds/ai-researcher works
Overall System Context
– The Research Agent layer, which handles user queries (at Level 1 and Level 2), orchestrates research tasks, and commands paper writing.
– The INNO engine, a module used by the research agent to execute the research workflow.
Main Components to Highlight
• INNO Core: Represent the core orchestration engine (folder “research_agent/inno/”) as a central component. This core handles the logic for research task management.
• Workflow: Optionally, include a “Workflow” component (research_agent/inno/workflow) that shows the dynamic flow—like flowgraph and flowcache—used to schedule and propagate research tasks.
Data Flow and Interactions
• Agents block to the Memory module, demonstrating that agents refer to stored code trees, previous paper drafts, and other past work.
• Tools to both INNO and Environment—since tools serve as the bridge to external APIs (e.g., GitHub, arXiv, and other research databases).
• Feedback loops: For example, upon experiment validation, results are stored back into the Memory module and then influence subsequent iterations.
– From user input → Research Agent → INNO.
– From Agents to Environment and Tools for accessing external data.