- [ ] Integrate Automatic Flame Graph Collection (useful to support in general) - [ ] Analyze it to look for bottleneck functions - [ ] Query an LLM to see if those bottleneck functions can be associated with particular tunable parameters - [ ] Focus some trials on those parameters in particular