Atharva Gupta

You can orchestrate end-to-end autonomous research projects, benchmark LLMs on 60+ academic suites, audit research artifacts for epistemic rigor across six scored dimensions, generate publication-quality matplotlib figures and architecture diagrams, and write camera-ready ML papers for NeurIPS, ICML, ICLR, ACL, AAAI, and COLM.

ML researchers, academic authors, and model evaluation engineers. The autoresearch and research-manager skills suit teams running multi-hypothesis experiments, while rigor-reviewer serves reviewers needing objective pre-publication critique, and ml-paper-writing targets authors preparing conference submissions.

The autoresearch skill runs a two-loop architecture: an inner loop executes rapid sandboxed experiments logging to results.tsv via eval-worker, while an outer loop synthesizes results and steers direction. Research-manager handles hypothesis planning, subagent routing, and approval gates, with continuous operation via Claude Code /loop or OpenClaw heartbeat.

Yes. All versioned skills in the manifest, including autoresearch, evaluating-llms-harness, rigor-reviewer, academic-plotting, brainstorming-research-ideas, and ml-paper-writing, are released under the MIT license by Orchestra Research at no cost.

LLM evaluation requires lm-eval, transformers, and vllm. Figure generation needs matplotlib>=3.8.0, seaborn>=0.13.0, numpy, and google-genai>=1.0.0. Paper writing requires semanticscholar, arxiv, habanero, and requests. The rigor-reviewer and brainstorming skills have no external dependencies.