What problem does it solve?
It converts a broad research direction into concrete, publishable research ideas and ranks them by feasibility, novelty risk, and expected impact.
Core Features & Use Cases
- Landscape survey: Scans local papers and searches recent literature to map sub-directions and identify explicit structural gaps.
- Idea generation + filtering: Brainstorms 8–12 ideas using an external LLM, then removes ideas that are infeasible, low-impact, or likely already done.
- Pilot validation: Runs short, cheap GPU pilot experiments for the top candidates to re-rank based on empirical signal, then writes a structured IDEA_REPORT.
- Optional wiki integration: If
research-wiki/ exists, writes idea pages and (when available) adds edges and rebuilds the query pack for memory and traceability.
Quick Start
Ask the agent to run the idea discovery pipeline for your topic and return a ranked report: "Generate and rank research ideas for: [your research direction]".