DSPy: The End of Prompt Engineering
summary
Claims that DSPy replaces brittle prompt tinkering by letting developers write typed Python programs with LLM calls as first‑class functions, using Signatures, Modules, Adapters, Optimizers to compose logic. Demonstrations include a file router that classifies SEC filings vs contracts, a PDF boundary detector that segments pages by visual layout, and an attachments-based RAG pipeline; optimizer runs (e.g., JEPA/MIPRO-style) lift a metric from ~86% to ~89% and produce a serializable optimized module that replaces hand‑tuned prompts.
AI-written summary · about
chapters (14)
- 0:00Introduction & The Enterprise AI Challenge
- 7:12The 6 Core Concepts of DSPy (Signatures, Modules, Adapters)
- 13:23Deep Dive: Class-based vs. Shorthand Signatures
- 19:57Adapters: Controlling the Prompt Format (JSON vs. BAML)
- 24:17Optimizers: The "Killer Feature" for Transferability
- 31:08Code Walkthrough: Setup & Model Mixing
- 36:24Handling Documents: "Poor Man's RAG" with Attachments
- 42:10Adapter Comparison: Improving Token Efficiency with BAML
- 47:20Optimizers in Practice: Creating Datasets & Metrics
- 51:13Complex Pipeline: Routing & Classifying Arbitrary Files
- 56:00Advanced Use Case: PDF Boundary Detection via Visuals
- 1:01:22Analyzing Optimization Results & The "DSPy Hub" Concept
- 1:09:02Q&A: Handling Delayed Feedback & Online Learning
- 1:13:00Conclusion
related talks
31mDSPy: The End of Prompt Engineering as We Know It
26mDSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines
19mOn Engineering AI Systems that Endure The Bitter Lesson
30mIs Prompt Engineering Dead?
30mIs Prompt Engineering Dead? How Auto-Optimization is Changing the Game
29mIs Prompt Engineering Dead? How Auto-Optimization is Changing the Game
39mOptimizing a Prompt for Production
13mPrompt Engineering Simplified