autoresearch-bootstrap

Automating ML research involves inspecting your repository, classifying files into fixed infrastructure and modifiable training code, then generating a Claude Code workflow to run autonomous experiments, log memory, and track results.

Separating code components uses heuristics to classify files into a fixed prepare.py scope for infrastructure and a modifiable train.py scope for training logic, ensuring safe autonomous edits by constraining modifications to the training scope.

Yes, you can generate an autonomous experimentation loop for any ML project by extracting the project name, typical training command, metrics, and baseline to produce an end-to-end workflow with structured tracking.

To start autonomous ML research, open Claude Code in your project directory and run the bootstrap command to automatically map data flow, extract metrics, and generate the experimentation workflow.

Autonomous ML research bootstrapping constrains edits strictly to the train.py scope, keeping prepare.py as fixed infrastructure to prevent unintended modifications to data preparation and configuration systems.

You need an existing ML codebase with identifiable entry points and configuration systems, allowing the bootstrapper to extract training commands and baselines to generate the automated workflow.