DSPyWeekly Issue No #2

Exploring GEPA and DSPy for AI system optimization

GEPA, or Genetic-Pareto, is a sample-efficient optimizer based on three principles: Genetic evolution, Pareto filtering, and Reflection using natural language feedback.

DSRs, @DSPyOSS for Rust

Happy to finally announce the stable release of DSRs. Over the past few months, I’ve been building DSRs with incredible support and contributions from folks Maguire Papay, @tech_optimist, and @joshmo_dev. A big shout out to @lateinteraction and @ChenMoneyQ who were the first people to hear my frequent rants on this!! Couldn't have done this without all of them.

@DSPyOSS module which lets you wrap another module and make it stateful - allowing multi-turn conversations

Created a little custom @DSPyOSS module which lets you wrap another module and make it stateful - allowing multi-turn conversations.

Manual Tool Handling

For more control over the tool calling process, you can manually handle tools using DSPy's tool types.

Where are DSPy users from?

GA analytics shared by DSPy team.

Context Engineering: Isaac Miller on Context Engineering with DSPy

Context engineering is rising in popularity because prompting alone isn't enough—we're still figuring out how to build reliable AI systems. From extracting structured data out of messy email threads to managing complex inputs like images and attachments, building useful LLM apps means more than just writing clever prompts. You need consistency, flexibility, and a way to reason about your system beyond a fragile block of text. Isaac Miller explains this through the lens of DSPy, a framework that turns prompting into programming. This talk was recorded at Chroma’s Context Engineering event in New York City on July 7th, 2025.

No metric? No problem. Optimizing Dspy programs with human in the loop

Getting started with Dspy can be difficult because we often don't see great performance out of the box without using an optimizer. | Channel: Skylar Payne

Context Engineering with DSPy - the fully hands-on Basics to Pro course!

This comprehensive guide to Context Engineering shows how to build powerful and reliable applications with Large Language Models (LLMs). I'll cover everything from atomic prompts and then use DSPy to build complex LLM systems. This includes RAG, tool calling, and multi-agent systems.

DSPy 3.0 Launch

The DSPy OSS team at Databricks and beyond is excited to present DSPy 3.0, targeted for release close to DAIS 2025. We will present what DSPy is and how it evolved over the past year. We will discuss greatly improved prompt optimization and finetuning/RL capabilities, improved productionization and observability via thorough and native integration with MLflow, and lessons from usage of DSPy in various Databricks R&D and professional services contexts.

Accelerate End-to-End Multi-Agents on Databricks and DSPy

Building a production-ready GenAI application requires much more than just a framework. Similar to traditional ML, you need a unified platform that supports the entire end-to-end workflow for production-grade applications. Here’s how this comes together on Databricks: Data ETL with DLT and Jobs Data storage, governance, and access with Unity Catalog Code development using Notebooks Agent versioning and metric tracking with MLflow and Unity Catalog Evaluation and optimization with Mosaic AI Agent Framework and DSPy Hosting and monitoring infrastructure via Model Serving and AI Gateway Front-end applications built with Databricks Apps In this session, you’ll learn how to: Build agents that can access all your data and models through function calling. Use DSPy to enable agent-to-agent collaboration, ensuring accurate answers.

sachink1729/DSPy-Multi-Hop-Chain-of-Thought-RAG

Discover advanced AI techniques in my repository combining Multi-Hop Chain of Thought (CoT) and Retrieval-Augmented Generation (RAG) using DSPy and Indexify. Enhance complex problem-solving with multi-step reasoning and external knowledge integration. Perfect for AI enthusiasts and researchers. | Language: Jupyter Notebook | Stats: 3 forks, 15 watchers | License: Apache License 2.0

dleerdefi/llm-security-auditor

DSPy-powered optimization highlights its unique strengths in LLM security by combining intelligent attack classification, dynamic analysis, automated prompt optimization, and professional tracking. It leverages LLM reasoning to classify jailbreak attempts with confidence scoring and detailed explanations, while adapting automatically to new attack patterns. Through DSPy modules, it performs dynamic vulnerability analysis, identifies risk levels, and provides contextual recommendations tailored to each prompt. Its optimization algorithms, including MIPROv2 and BootstrapFewShot, enhance security without sacrificing functionality—demonstrated by reducing jailbreak rates from 23% to 3%. With seamless MLflow integration, progress tracking, and rich reporting, DSPy showcases its ability to deliver production-ready security solutions for LLM applications. | Language: Python | Stats: 3 forks, 20 watchers | License: MIT License

ganarajpr/awesome-dspy

An Awesome list of curated DSPy resources. | Stats: 25 forks, 424 watchers

sachink1729/SQL-Agents-Using-RAG-DSPy-Groq

Exploring advanced prompting tools to query SQL database with multiple tables in natural language using LLMs | Language: Jupyter Notebook | Stats: 3 forks, 13 watchers | License: Apache License 2.0

ganarajpr/sanskrit-translator-dspy

Using DSPy and LLM's to translate Sanskrit verses | Language: Python | Stats: 2 forks, 12 watchers

KarelDO/xmc.dspy

Infer-Retrieve-Rank (IReRa) is a generic and modular program which specifies interactions between pretrained Language Models and Retrievers to efficiently solve multi-label classification tasks with an extreme amount of classes (≥10,000 classes). Using only ≅ 50 labeled input examples, IReRa can be optimized to achieve state-of-the-art performance, despite not being finetuned. This optimization generally involves having a strong, teacher Language Model (e.g. gpt4) solve the task and gather instructions or demonstrations that help a more efficient, student Language Model (e.g. llama-2) solve the task better. A user can easily specify which parts of the program are implemented using which LMs, to strike the perfect balance between cost and performance. | Language: Python | Stats: 24 forks, 6 open issues, 435 watchers | License: MIT License

XiaoConstantine/maestro

Maestro is an intelligent code review assistant built with DSPy-Go that provides comprehensive, file-level code analysis for GitHub pull requests. It combines advanced AST parsing, semantic analysis, and LLM-powered reasoning to deliver high-quality, actionable code review feedback. Additionally, Maestro features seamless integration with Claude Code and Gemini CLI tools for enhanced AI-powered development workflows. | Language: Go | Stats: 5 open issues, 10 watchers | License: MIT License

kuzudb/dspy-kuzu-demo

Graph data enrichment using DSPy and Kuzu - This repo contains an example of using DSPy for graph data enrichment, i.e., enriching data from one source with data from another source. We'll do this by reframing the problem as an entity disambiguation task, where we match entities from the two datasets based on their attributes using vector search, and then merge the data from the richer of the two sources, to create enriched data that we can then use to build a more useful knowledge graph in Kuzu. | Language: Python | Stats: 11 watchers | License: MIT License