Propel: Structured Constraint Framework for Claude Code in Research Workflows

📅 Project Timeline: Feb 2026 - Now

Propel is a structured constraint framework designed to enhance Claude Code's effectiveness in research workflows. Unconstrained LLMs tend to produce "the mean of training data" rather than the specific, precise outputs that research demands. Propel addresses this by introducing human-in-the-loop gates, domain-specific auditors, and an investigation-first workflow that ensures Claude's outputs match specific research requirements.

Propel framework powered by Claude
Propel framework powered by Claude Code.

Key features include five human-in-the-loop checkpoint gates, domain-specific auditors for verifying code against research papers and tracing tensor shapes through JAX transforms, a session management system for organizing research artifacts, and customizable skills for literature review, design planning, and validation. This website itself was built using Propel with Claude.

GitHub Repository