NeuriCo

Train or fine-tune models on Modal cloud GPUs (LoRA SFT, full SFT, data prep, eval) with strict leave-no-trace lifecycle. Use when an experiment needs GPU training, fine-tuning, or model evaluation that exceeds local Docker capacity.

Serve a model via vLLM on Modal as an HTTPS endpoint with strict leave-no-trace lifecycle. Use when an experiment needs to query an LLM as a remote API (chat completions, generation) rather than running it locally.

Use the University of Chicago Data Science Institute Slurm cluster, called dsi-cluster here, for NeuriCo training, evaluation, sweeps, and batch jobs. Use only when NeuriCo is run with --compute-backend dsi-slurm. Run jobs in the runtime-provided remote workspace, monitor Slurm safely, and copy required outputs back to the local NeuriCo workspace.

Formally verify Lean 4 proofs against Mathlib using the hosted Axiom AXLE API — no local Lean toolchain. Use in the mathematics_lean domain, when a proof needs machine-checked verification but installing Lean/Mathlib locally is undesirable, or when translating informal proofs into formally verified Lean 4 code checked in the cloud.

Autonomous research framework that orchestrates AI agents (Claude Code, Codex, Gemini) to design, execute, analyze, and document scientific experiments. Takes a structured research idea (YAML with title, domain, hypothesis) and produces code, results, plots, LaTeX papers, and GitHub repositories.

Write academic papers from experiment results using LaTeX. Use when experiments are complete and REPORT.md exists, when asked to write a paper, or when generating publication-ready documents in NeurIPS style.

Find and search for academic papers using the paper-finder service. Use when conducting literature review, searching for related work, finding baseline papers, or looking for methodology references.

Conduct systematic literature reviews using a structured workflow. Use when starting a new research project, surveying a field, or documenting related work for a paper.

Manage BibTeX citations including DOI-to-BibTeX conversion, metadata extraction, validation, and deduplication. Use when building reference lists, converting DOIs to citations, or validating existing .bib files.

Evaluate research quality using frameworks like GRADE, bias detection, and logical analysis. Use when assessing paper quality, evaluating claims, or reviewing evidence strength.

Conduct exploratory data analysis and statistical testing with test selection guidance. Use when exploring datasets, selecting statistical tests, performing power analysis, or preparing results for publication.

Write research grant proposals with guidance for NSF, NIH, DARPA, and other agencies. Use when preparing grant applications, developing research plans, or writing broader impacts statements.

Formulate research hypotheses using structured frameworks. Use when developing research questions, designing experiments, or planning studies with testable predictions.

Conduct structured manuscript reviews with section-by-section checklists and constructive feedback. Use when reviewing papers before submission, providing feedback to colleagues, or preparing reviewer responses.

Creative ideation for research using structured methods like SCAMPER, morphological analysis, and cross-domain analogies. Use when generating research ideas, exploring new directions, or overcoming creative blocks.