Home | Weekly Newsletter

Learning DSPy (3): Working with optimizers • The Data Quarry

Welcome back to the Learning DSPy series! So far, we’ve discussed the core building blocks of DSPy (signatures and modules, without going into optimizers), and highlighted the strengths of the programming model offered by DSPy. In this post, we’ll take a closer look at optimizers, and how they can be used to automatically improve the performance of a DSPy program.

🎧 He Built an AI Ghostwriter With Taste

How Danny used DSPy to give AI taste: 00:47:52

Paper - SAVANT: Semantic Analysis with Vision-Augmented Anomaly deTection

Researchers introduce SAVANT, a structured reasoning framework that enhances anomaly detection in driving scenes using vision-language models (VLMs). Unlike ad-hoc prompting, SAVANT analyzes images across four semantic layers — Street, Infrastructure, Movable Objects, and Environment — achieving over 90% recall and accuracy with open-source models like Qwen2.5VL. This approach not only boosts reliability but also enables low-cost, local deployment and automatic labeling of thousands of real-world images, tackling the long-tail challenge in autonomous driving. Makes use of DSPy.

Paper - MTP: A Meaning-Typed Language Abstraction for AI-Integrated Programming

This paper presents Meaning-Typed Programming (MTP), a novel paradigm that abstracts LLM integration through intuitive language-level constructs. By leveraging the inherent semantic richness of code, MTP automates prompt generation and response handling without additional developer effort. References and talks about DSPy.

How to build your own long-term Agentic Memory System for LLMs | Mem0 from scratch w DSPy

In this video we are using DSPy and QDrant Vector Database to create our own memory system from scratch! We are building the core components of Mem0 from the ground up, one step at a time. You'll learn about the Mem0 API, the basics of DSPy signatures and modules, generating embeddings, inserting and searching with vector databases (QDrant), and tool calling with dspy.React.

Logic Meets Automation

DSPy boosted GPT-3.5 Accuracy by 19.5%.

Using Tools (ReAct Pattern) with ax-llm & DSPy

Let your AI use tools to answer questions - the ReAct (Reasoning + Acting) pattern in ax-llm, DSPy

Meaning-Typed Programming: Types as Prompts for LLMs - YouTube

In this walkthrough I break down the SPLASH 2024 paper on meaning-typed programming and show how its type-driven approach can remove a ton of brittle prompt wiring from our code. Paper: https://arxiv.org/pdf/2405.08965 Resources Meaning-Typed Programming (Jack language) paper SPLASH 2024 conference proceedings

DSPy - Testimonial by usage

Convergence FM Podcast

Route to Success: Scalable Routing Agents With Databricks and DSPy

The talk explores how to manage multiple AI assistants through a **flexible Routing Agent** that automatically directs queries to the right model. Using **Databricks** and **DSPy 3.0**, it covers optimizing routing accuracy, reducing latency, enabling stateful and scalable interactions, enforcing access control, and tracking performance in production. Attendees will learn practical strategies to build efficient, adaptable multi-assistant systems.