Jayita Bhattacharyya: DSPy the Declarative Programming in the Era of AI

See https://dspy.ai

DSPy is an open source declarative framework for building modular, self-improving AI systems using structured code and natural-language modules, enabling fast iteration, composability, and model-agnostic deployment.

Prompts are a great way of communicating with LLMs, but they need to be elaborate and exhaustive, hard to maintain when changing from one LLM to another.

DSPy: the vibe prompter via programming. Small demo. You can ask it to show the prompt that is has actually sent to the LLM. https://hamel.dev/blog/posts/prompt/

GEPA: the game changer. Generic Pareto. It is a framework for optimizing arbitrary systems composes of text components, like AI prompts, code snippets, or textual specs, against any evaluation metric. It employes LLMs to reflect on system behavior and user feedback.

[Interesting talk, but not my area of expertise and way too much to write down, so look at the slides.]