Which reasoning approaches do you use often?

In my LLM pipelines, I've found myself doing the following: 1. CoT and prompt chains - - the highest ROI approach, breaking down the task into smaller reasoning tasks and giving the model a rough structure of thought and some examples. I'll add CoT+Reflection as a subset of this approach. 2. Best of N Sampling using a second "judge" model 3. Self Consistency - - clustering responses and picking the representative response from each cluster...this reduces the error rate by preventing those one-off poor quality responses from being used OptiLLM has a comprehensive set of SoTA reasoning: https://github.com/codelion/optillm I'm especially curious if anyone is using PlanSearch, R*, MCTS and other search based approaches, and what your use case is.