Stay Updated with DSPyWeekly!
DSPyWeekly Issue No #12
Published on November 21, 2025
đź“š Articles
Announcing DSCloj - By Kapil Reddy
Announcing DSCloj! A declarative way to do prompt engineering in Clojure. It is inspired by DSPy library in Python. In it’s current shape API looks very similar to instructor-clj right now. But next up DSCloj will have optimisers too.
DSPy on a Pi: Cheap Prompt Optimization with GEPA and Qwen3
Lee Butterman’s recent experiment, "DSPy on a Pi," demonstrates that sophisticated prompt optimization is achievable on low-cost edge hardware like a Raspberry Pi 5 rather than expensive GPU clusters. By utilizing the DSPy framework and Generative Evolutionary Prompt Adaptation (GEPA), Butterman tasked a mid-sized "teacher" model (Qwen3 4B) with critiquing and rewriting prompts for a tiny Qwen3 0.6B model. Over the course of 16 hours, this iterative process improved the smaller model's ability to translate natural language into SQL from a meager 7.3% baseline to a 28.5% success rate, highlighting the growing viability of powerful AI optimization on accessible, low-power devices.
Omar Khattab on X - Code Snippet
A cool thread yesterday used GPT4 ($50), a 500-word ReAct prompt, and ~400 lines of code to finetune Llama2-7B to get 26% HotPotQA EM. Let's use 30 lines of DSPy—without any hand-written prompts or any calls to OpenAI ($0)—to teach a 9x smaller T5 (770M) model to get 39% EM!
What TOON Gets That CSV Doesn’t for LLM Payloads
Token-Oriented Object Notation keeps your nested Sorbet structs intact—something flat CSV rows simply can’t do when you prompt large language models. By Vicente Reig Rincon de Arellano of DSPy.rb
Building Intelligence Systems with DSPy, MCP and Mem0 (Part 2) | by Jitendra Jangid | Nov, 2025 | Medium
Adding Memory, Production UI, and Multi-Step Reasoning How we evolved from a stateless prototype to a production-ready intelligence system with conversation memory.
Twitter thread on FastMCP basics and DSPy mcp host+client implementation
Show a video on integrated output with claude desktop.
Accelerating data modeling accuracy with the Amazon DynamoDB Data Model Validation Tool | AWS Database Blog
DSPy used alongside Amazon Bedrock, Strands Agents.
Naive Text-to-SQL Implementation
In this article, we’ll understand what DSPy means and build a practical Text-to-SQL application to see how it works.
PAPER - AISAC: An Integrated multi-agent System for Transparent, Retrieval-Grounded Scientific Assistance
Developed by researchers at Argonne National Laboratory, AISAC (AI Scientific Assistant Core) is a specialized multi-agent framework designed to bring transparency and reliability to scientific workflows. Built upon LangGraph (orchestration), FAISS (vector search), and SQLite (persistence), the system uses a Router-Planner-Coordinator architecture to decompose complex research problems while rigorously tracking the "provenance" of every tool output and decision. DSPy is cited in the Related Work as a prominent framework for programming language models,
PAPER - IndicGEC: Powerful Models, or a Measurement Mirage?
This technical report by Sowmya Vajjala (TeamNRC) details experiments conducted for the BHASHA-Task 1, a shared task focused on Grammatical Error Correction (GEC) for five Indian languages: Hindi, Telugu, Tamil, Malayalam, and Bangla. DSPy implementation in the code.
Alex Noonan on X - Code snippet
For the AI component or so-called "Agent", I used dspy which has the best abstractions for building AI based software these days, very pythonic and makes your code feel alot of more composable and structured.
🎥 Video
Why are prompt optimizers still so underrated?
By Chris Potts (Stanford Professor ) Event: Bay Area DSPy Meetup (Nov 18, 2025) ... We wouldn't dream of manually setting the weights of a neural network by hand, yet we spend hours "fiddling" with prompt strings hoping for a better output. In this talk, Chris Potts argues that the era of manual prompt engineering is ending—and why automated optimization is the missing link for reliable AI systems.
AIdotEngineer Impromptu conversation around DSPy
Found the DSPy / GEPA corner at the Anthropic booth with @tarunsachdeva and @thomastjoshi
The era of self-learning AI – Prompt optimisation with DSPy
In Japanese language, introduction to DSPy.
Youtube Reel - GEPA Introduction (Genetic-Pareto Prompt Optimizer)
GEPA can improve your AI agent performance simply by evolving the instruction or prompt.
🚀 Projects
lsb/dspy-sql-chat
Using DSPy to optimize Chat-to-SQL | Language: Python | License: GNU Affero General Public License v3.0
large-scale-ai-systems/incschema-Dspy
Language: Python | License: MIT License
marcusjihansson/trading-researcher-in-go
This is a trading researcher built in Golang, which analyzes articles and outputs a summary. Everything iwth Gemini api and dspy. | Language: Go
Thanks for reading DSPyWeekly!