DSPy Integration — Use TruLens Feedback as DSPy Metrics

[joshreini1]

Summary

DSPy is the most popular prompt/pipeline optimizer. TruLens feedback functions are a natural fit as DSPy metrics — users could use TruLens evals (groundedness, relevance, etc.) to drive DSPy's BootstrapFewShotWithRandomSearch, MIPROv2, or BayesianSignatureOptimizer. Today there's no bridge; users have to manually wrap TruLens feedback functions as DSPy metric callables.

What

Create a trulens-integrations-dspy package (or a module in src/feedback/) that provides:

• TruLensMetric(feedback_fn) — a wrapper that adapts any TruLens feedback function into a DSPy-compatible metric (accepts example, prediction and returns a float)
• A cookbook showing: define a DSPy module, use TruLens groundedness + relevance as optimization metrics, run BootstrapFewShotWithRandomSearch, compare before/after with the TruLens dashboard
• Optional: record DSPy optimization traces as TruLens records so users can see which prompt variants were tried

Difficulty

Medium