StaRPO: Stability-Augmented Reinforcement Policy Optimization

Build an RL policy optimizer that rewards internal structural consistency rather than just output correctness. Develop a custom loss function that penalizes illogical reasoning paths.

Suggested repo: logic-rl

"Train LLMs that think, don't just guess."

Estimated effort: 90h