Unlocking DSPy for Smarter LLM Applications

From Prompt Engineering to AI Programming – Unlocking DSPy for Smarter LLM Applications

Our April Meetup will be led by Chen Qian, a senior software engineer at Databricks responsible for building and maintaining DSPy. DSPy is an exciting new tool that turns today's prompt guesswork into an engineer-friendly and automated approach. This month our event will at the Proud Bird - near LAX.

Come at 6:30pm to socialize with our growing group. Beer, wine and appetizers will be served. Presentation will start at 7pm.

Talk Details:

Part 1: The Rise of Compound AI Systems & Why We Need DSPy (5 min)

• The shift from basic chatbot to compound AI systems (agents) in response to generative AI advancements.
• Existing tools like LangChain & LlamaIndex—and why many developers end up writing custom solutions instead.
• Introducing DSPy: a framework that integrates easily into existing workflows, enabling structured LLM programming instead of ad-hoc prompting.

Part 2: DSPy’s Key Offerings – How It Transforms AI Development (15 min)

• Programming vs. Prompting
• How DSPy replaces long, hand-tuned prompts with structured programming using Signatures & Adapters.
• Automatic Prompt Optimization: How DSPy optimizers does automatic prompt optimization, we won’t dive too deep into this, but do a brief explanation.
• How to choose the right optimizer for different tasks.
• Inference-Time Optimization: dspy.Refine and other techniques that allow self-improvement at runtime using additional signals.

Part 3: Productionizing DSPy (10min)

• Bringing DSPy programs to production with MLflow
• How MLflow tracing helps track, iterate, and optimize DSPy programs
• How to iterate through and track your experiments with MLflow.
• Deploying DSPy programs with MLflow, including streaming support for real-time applications.

Part 4: Question & Answering (10min)