CoolFace
Apppublic

coredipper/operon-chaperone

sourceHugging Facemitupdated 6mo agoView on Hugging Face
1likes
App README

🧬 Operon Chaperone

Recover structured data from malformed LLM output using a multi-strategy cascade -- like protein chaperones refolding misfolded proteins.

What to Try

  1. 1.Open the Input & Parse tab, select a schema from the Target Schema dropdown (e.g. FunctionCall), paste broken JSON into the input box, and click Parse to watch the cascade recover it.
  2. 2.Toggle individual strategies on or off (STRICT, EXTRACTION, LENIENT, REPAIR) to see which one succeeds for different kinds of malformation.
  3. 3.Switch to the Batch Examples tab and click Run All Examples to see the cascade handle trailing commas, single quotes, markdown-wrapped JSON, and more in one pass.

How It Works

The Chaperone tries four folding strategies in order -- strict parsing, extraction from wrappers, lenient type coercion, and structural repair -- stopping at the first success. This mirrors how biological chaperones apply progressively stronger refolding forces to recover functional protein structure.

Learn More

GitHub | PyPI | Paper