Cadenza bridges Wandb logs to autonomous AI agents

Cadenza solves the "context rot" problem for AI researchers by treating ML experiments as an evolutionary search space rather than a flat list of logs.

• –Transforms Weights & Biases projects into structured "Genotypes" and "Islands" that AI agents can actually reason about
• –Provides a "sample" command that generates JSON snapshots optimized for LLM context windows, reducing token waste
• –Enables a closed-loop "Sample → Decide → Execute → Import" workflow for fully autonomous machine learning research
• –Lightweight Python SDK (only httpx and pydantic) makes integration into existing agentic workflows painless
• –Encourages diverse exploration through subpopulation tracking, preventing agents from getting stuck in local optima