DSPy and GEPA Optimization

Hand‑writing prompts is hard so why not use DSPy instead? This session shows how to turn LLM work from prompt tinkering into structured, optimizable programs. It introduces DSPy’s core abstractions—signatures through typed input–output contracts, modules through composable building blocks, and optimization through selecting examples, synthesizing prompts, and tuning module parameters. A brief tour of the GEPA optimization explains how scoring and feedback generates optimized prompts.

During the session we'll build a small pipeline end‑to‑end: define a signature, wire up a module, and establish a baseline then capture runs as dspy.Example objects, curate them into gold examples, and grow the set in order to run GEPA, generating feedback through an LLM-as-judge that mimics the type of feedback we care about.

Join us for the next DSM AI Group!

5:30 - 6:00 - Socializing/Networking
6:00 - Speaker begins