DSPyWeekly Issue No #5

Context Engineering: Improving AI Coding agents using DSPy GEPA

This blog post is a technical walkthrough of how we improved the coding agents used in the AI data scientist. With the actual data, and evaluation technique explained. It has the following portions: Preparing data Explaining GEPA Applying prompt optimization (GEPA) via DSPy Results

Cache Image.format for better throughput (#8842) - DSPy code commits

lru_cache cache decorator applied to ensure faster throughput and stopping needless regeneration.

Minor SIMBA Improvements - From DSPy codebase

Added ability to use a seperate prompt_model to use to evolve the program.

Tracking DSPy Optimizers

This tutorial demonstrates how to use MLflow to track and analyze your DSPy optimization process. MLflow's built-in integration for DSPy provides traceability and debuggability for your DSPy optimization experience. It allows you to understand the intermediate trials during the optimization, store the optimized program and its results, and provides observability into your program execution.

Using DSPy in Production

The following content is not an introduction to DSPy, nor is it a tutorial to learn how to use DSPy. I believe this topic has already been well covered (see articles by Maxime Rivest, DSPy-0-to-1 and the DSPy Documentation). However, I believe that there’s a lack of content on how to bring an AI-based app using DSPy to production, notably on using Async DSPy and I aim to close part of this gap with this article, inspired by a project that I’ve been working on.

DSPyWeekly Updates

A lot of DSPy content is crawled compared to what is curated, now you can search it at https://dspyweekly.com/search/ . A new section for Companies that use DSPy added to the newsletter. Check the section at the end of the newsletter. Use https://dspyweekly.com/submit/ to send in any relevant link you will like for curation in the newsletter.

Prompt Engineering 的終結？DSPy 框架概念 + n8n 模擬測試，讓 AI 自動寫出最佳提示詞！ - YouTube

DSPy quick review in Chinese. 隨著模型性能提升， Prompt Engineering 的重要性不如數年前那麼顯著。今年一個非常受歡迎的項目 DSPy 提出了一個革新性的概念：「Programming—not prompting」，主張讓程式自動生成 Prompt、測試和選擇最適合的 Prompt，不再需要人手寫。在這影片，我透過 n8n 模擬並講解當中的原理。

DSPy Primer: Mike Taylor Session With Every.to

Listen to this introduction to DSPY by Micheal Taylor, a key architecture underpinning AskRally's infrastructure choices. | Channel: Rally

Image extraction with DSPY | Python | Gemini

How to extract any information from images using DSPY and Gemini models. | Channel: kevin herbas

Building AI Agents with Gemini & DSPy

This project was created for a live coding session where we explored building AI bots and agents using Google’s Gemini 2.5 Flash-Lite and DSPy. In the session, we walked through integrating the Gemini API into applications, understanding how DSPy simplifies working with LLMs and prompt engineering, and building a practical AI agent from scratch that can perform real tasks. We also addressed common issues and debugging challenges as they came up, making it a fully hands-on experience.

CTO update: The DSPY framework to automate and control LLM behavoir

Hear from the CTO of Algorithma on his view of DSPy.

GEPA: A New LLM Prompt Optimizer Beats RL

AI generated video on GEPA Paper.

langstruct-ai/langstruct

Extract structured data from any content using LLMs. | Language: Python | License: MIT License

S1M0N38/dspy-arxiv

Explore the use of DSPy for extracting features from PDFs 🔎 | Language: HTML | License: MIT License

jmanhype/dspy-self-discover-framework

Leveraging DSPy for AI-driven task understanding and solution generation, the Self-Discover Framework automates problem-solving through reasoning and code generation. | Language: Python

StanfordMIMI/dspy-helm

Toward Reliable, Holistic Evaluation of Language Models with Prompt Optimization | Language: Python | License: MIT License

Sckathach/dspy-tap

This project is a short implementation of the Tree of Attacks (TAP) attack with the DSPy framework. | Language: Python

jmanhype/MemesPy

🎭 DSPy Meme Generator: AI-powered meme creation using DSPy and FastAPI. Generate hilarious, relevant memes with intelligent text and image generation. Ready for GPT-4o's upcoming image capabilities! 🚀 | Language: Python

BoxiYu/DSPy-Guardrails

Building self-refined guardrails via DSPy | Language: Jupyter Notebook

Sruthi-Pedakolimi/MarketMind

AI-powered stock analysis platform with real-time data integration and natural language processing. Built with React, FastAPI, and DSPy for intelligent financial insights. | Language: TypeScript

wangjing0/gepa-optimizer

GEPA: Genetic-Pareto Evolutionary Algorithm for Prompt Optimization | Language: Python

rsrini7/RAG-Hybrid-Inference-Pipeline

This application demonstrates a Retrieval-Augmented Generation (RAG) pipeline built with DSPy & inference pipeine using LitServe. It combines keyword search (BM25) and dense vector search (ChromaDB with Sentence Transformers) followed by a reranking step (Cross-Encoder) to retrieve relevant context for a Language Model (LLM) to generate an answer. | Language: Python

baskar-ak/Agentic_RAG

An Advanced Multi-hop Agentic RAG system. ReAct Agent & MLflow Tracking. Tools: Pinecone, DSPy, ChatGPT, MLflow, Langchain, Tavily. | Language: Python

Modaic - Placing DSPy within the Software 3.0 ecosystem

Modaic is the platform for DSOps: DevOps for DSPy programs. We’re building the infrastructure for software 3.0, where agents autonomously improve through structured feedback loops. Our SDK and hub enable developers and enterprises to rapidly build specialised agents from first principles, providing primitives that streamline the develop, deploy, evaluate, and optimise cycle.