Context Engineering with DSPy

Chapter 3. DSPy in 8 Steps

A note for Early Release readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 3rd chapter of the final book. The GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

Now you know the individual components of DSPy programs, it’s time to build and optimize your first DSPy program end-to-end. We’ll work through the step-by-step process I use for every DSPy program, from creating Signatures and Modules, to formalizing your data and evaluation metrics, then testing and improving your program. You’ll get to see the magic of DSPy’s declarative framework, and how powerful its Optimizers are once you build your program in the right format.

We’ll see a huge performance gain on an exemplary use-case: rewriting AI-generated text to sound more human. We’ll employ the versatile LLM-as-a-Judge technique, training a DSPy program that detects AI-generated writing. We’ll use the judge to train another DSPy program that tries to “trick” the judge, rewriting AI-generated text into natural, human-like writing. This evaluator-optimizer pattern is something I use regularly to handle “fuzzy” creative tasks where it is difficult to develop formal evaluation metrics.