DSPyWeekly Issue No #16

Event: Beograd - Structured Context Engineering: Optimizing AI Agents with DSPy

Data Sanity Workshop: Optimizing AI Agent Prompts with DSPy In this hands-on workshop, you’ll be introduced to DSPy, a practical framework for optimizing prompts for AI agents.

Stop Writing Prompts Like a Medieval Alchemist | by DrSwarnenduAI | Dec, 2025 | Stackademic

The article critiques the inefficient "alchemy" of manual prompt engineering, advocating instead for the DSPy framework to transform LLM interactions into structured, programmable modules that automatically optimize themselves for consistent, reliable performance.

PAPER: The Meta-Prompting Protocol: Orchestrating LLMs via Adversarial Feedback Loops

The Meta-Prompting Protocol (MPP) replaces static prompting with an "Adversarial Trinity" (Generator, Auditor, Optimizer) that creates dynamic feedback loops. This game-theoretic approach enables LLMs to autonomously critique and refine their own instructions for improved reasoning and robustness without human intervention. MPP is highly relevant to DSPy as a potential "Auditor-based Teleprompter," offering a sophisticated optimization strategy that uses adversarial critique rather than just simple metrics. This could enhance DSPy's existing optimizers (like BootstrapFewShot) by helping pipelines actively detect errors and prevent overfitting.

Anton Lobach on DSPy - X

Why Anton prefers DSPy

How DSPy Builds Prompts?.

It demonstrates how DSPy automatically translates a Python class (defined with input/output fields and a docstring) into a structured system prompt that includes field markers and specific formatting instructions. By inspecting the model's history, the author reveals that DSPy ensures predictable, type-safe outputs (such as specific literals or floats) by explicitly guiding the LLM on how to map inputs to outputs, effectively bridging the gap between flexible LLM generation and reliable, structured programming.

Vicente Reig — DSPy.rb: a Ruby-first port of Stanford’s DSPy

Vicente Reig creator of DSPy.rb talking at ruby conference on DSPy.

Prompt Optimization with DSPy

As model performance converges, prompt optimization is the new competitive edge. In this session, we revisit DSPy to cover the state of prompt optimization, agent harness optimization, and the latest advances in evaluation-driven systems—including a deep dive into the GEPA (Genetic-Pareto) optimizer. Learn how to program, evaluate, and optimize AI agents with DSPy, and where it fits in a modern AI engineer’s toolkit.

DSPy to DSRs: Conversation with DSPy Committer Herumb Shandilya

In this conversation, Herumb Shandilya, a core maintainer of DSPy and developer of its Rust variant DSRS, discusses his journey with DSPy, the Rust port, and the future of LLM optimization.

sotayamashita/dspy-acp

Experimental DSPy adapter for the Agent Client Protocol (ACP) | Language: Python

Manojkumar2806/QdrantRAG

MedSage is a multimodal healthcare assistant that combines LLMs, vector search, and real-time reasoning to deliver fast, reliable medical insights. It supports symptom analysis, medical document Q&A, universal file RAG, multilingual interactions, and emergency SOS with live location. | Language: TypeScript

OniricApps/aisha

Aisha is a personal shopping assistant chatbot that helps users find products on Amazon and other online stores. Built with Flask and powered by Google's Gemini AI (via DSPy), Aisha provides conversational product recommendations, gift ideas, and personalized shopping assistance in Spanish. An OniricApps production. | Language: Python | License: GNU General Public License v3.0

Uzair-1006/Medsage-dspy

Language: Python

Ben (no treats) on X about not liking DSPy

Worth knowing why someone wouldn't like DSPy and see the conversation underneath.

Stay Updated with DSPyWeekly!