research-pipeline

The research pipeline automates the entire research lifecycle from idea discovery to paper writing. By inputting a research direction and configuring parameters like auto-proceed and venue, the workflow handles idea generation, experiment deployment, automated reviews, and paper writing sequentially.

The research pipeline includes an experiment bridge for automating code implementation and deployment, alongside an auto review loop to conduct automated code reviews and iterate on experiments. You can enable code review and configure human checkpoints during deployment.

The research pipeline requires dependencies including codex tools and deterministic experiment bridges to function. These AI models and tools facilitate different stages of the research process, from generating ideas to writing the final paper.

Yes, the paper writing feature supports targeting specific conference venues like ICLR. You can specify the venue parameter in the start command to generate a narrative report and format the paper submission accordingly.

You can start the research automation process by executing the start command with your research direction and setting the base repo parameter. This initiates the pipeline to generate and validate research ideas based on the provided repository context.

You should disable human checkpoints by setting the parameter to false when you want the pipeline to run fully autonomously without manual intervention. This allows the automated review loop to iterate on experiments and write papers continuously without pausing for manual approval.