OpenNSWM-Lab/FAROS

What it solves

FAROS addresses the lack of grounding and traceability in AI-generated research. It prevents LLMs from generating baseless ideas by requiring literature evidence first, replaces vague natural language handoffs between modules with structured contracts, and ensures that automated reviews are actionable by linking claims directly to evidence and experimental results.

How it works

FAROS operates as a multi-agent collaborative system that follows a structured scientific loop:

1. Idea Generation: Uses a "topic coach" to refine research interests into concrete tasks and retrieves literature that must pass quality gates before supporting an idea.
2. PlanPackage: Creates a structured contract containing hypotheses, variables, and acceptance criteria to guide downstream modules.
3. Code & Experiment: Generates runnable code in a sandbox, executes experiments, and collects structured metrics and figures.
4. Paper: Drafts a paper linked to the original plan and experimental results.
5. ReviewX: An auditing system that checks the consistency between claims, evidence, and measurements, feeding results back into the research plan for revision.

Who it’s for

It is designed for researchers and developers working on LLM and AI Scientist research who need a system that maintains a rigorous, auditable trail from the initial research question to the final paper.

Highlights

• Evidence-First Approach: Literature must be semantically aligned and validated before it can be used to justify a research idea.
• PlanPackage Contract: Uses a typed handoff system to ensure the code and review modules understand the exact scientific intent.
• ReviewX Closed-Loop: Transforms peer review from a simple score into a consistency audit that triggers specific research revisions.
• Human-in-the-Loop: Maintains human decision gates at critical points like plan approval and final sign-off.
• Recoverable Tasks: Supports long-running background jobs that survive browser refreshes or disconnects.