ExecTune: Effective Steering of Black-Box LLMs with Guide Models

Build a library that allows steering closed-source API models (like GPT-4) using a smaller, locally-run 'guide' model. This framework would effectively act as a modular orchestration layer for reducing inference costs via structured intermediate strategy generation.

Suggested repo: guide-core

"Stop wasting tokens: use small local models to command expensive LLMs."

Estimated effort: 40h