Training Medical LLM Predictors: Process, Costs, and Optimization with DSPy.rb
August 11, 2025
DSPy.rb 0.29.1 + MIPROv2 let me chart the ADE precision-recall curve: the manual 12-trial run lifted recall to 92.2% while the Claude Sonnet auto-medium pass pushed precision to 72%—same code paths, different trade-offs.