EvoMap/AutoResearch

AI/ML research agents from idea to paper-ready evidence. An EvoMap open-source project.

What it solves

AutoResearch automates the end-to-end process of AI and machine learning research, moving from an initial idea to a package of evidence ready for paper writing. It specifically targets the problem of "hallucinations" in AI research agents by grounding discovery in real-world signals, enforcing multi-model reviews, and maintaining a traceable, stateful record of all experiments, including negative results.

How it works

The system operates through two primary workflows:

  1. Idea Generation: It collects research signals from public channels (papers, community trends), intersects them with a local knowledge base of domain-specific constraints and experience, and uses at least three distinct models to independently develop and cross-review candidate ideas.
  2. Idea Execution: Using the Claude Code CLI and a coordinator, it manages a recoverable project lifecycle. This includes initializing a plan, running a small-scale pilot to test feasibility, scaling to full experiments, and performing independent blind reviews via a critic to challenge conclusions.

All progress—including code, logs, metrics, and critic reports—is persisted to disk, allowing researchers to inspect the process or take over manually.

Who it’s for

AI and machine learning researchers who want to automate the discovery of novel research directions and the execution of experiments while maintaining rigorous evidence and traceability.

Highlights

  • Cross-Domain Discovery: Combines external research signals with internal domain knowledge to find novel problems.
  • Multi-Model Validation: Requires distinct model identities for idea generation and review to prevent self-approval bias.
  • Stateful & Recoverable: Persists all work units, allowing long-running experiments to resume after interruptions.
  • Pilot-First Approach: Validates feasibility at a lower cost before committing to full-scale execution.
  • Evidence-Based: Records full provenance from the initial signal to the final blind review, explicitly preserving negative results.

Related

  • Project
    synthetic-sciences/openscience

    OpenScience is an open‑source AI research workbench that lets users describe scientific tasks in plain language. The built‑in agent plans, searches literature, writes and runs code (Python/R/shell), executes experiments on local or remote compute, and produces a reproducible report, with every step logged. Available as a desktop app, web workspace, or CLI, it supports custom LLMs, a large library of domain‑specific skills, and extensible plugins.

  • Project
    appautomaton/latex-arxiv-SKILL

    An agentic harness for writing verified AI/ML review papers in LaTeX, featuring a gated, issue-driven workflow that prevents citation hallucinations.

  • Project
    federicodeponte/opendraft

    OpenDraft is an open‑source Python engine that generates long‑form academic drafts with citations verified against multiple scholarly databases. It uses 19 specialised AI agents, supports 57+ languages, and exports to PDF, Word or LaTeX. Designed for researchers, students and developers, it provides citation‑checking, TL‑DR summaries, audio digests and data‑fetching utilities, but still requires human review before final publication.

  • Project
    WILLOSCAR/research-units-pipeline-skills

    A framework for agent-assisted research that transforms research goals into verifiable, resumable runs with a full evidence trail and pinned pipeline contracts.