Geek Out Time: From Prompts to Programs -How Open-Source DSPy + DeepSeek Train Better Autogen…

Geek Out Time: From Prompts to Programs -How Open-Source DSPy + DeepSeek Train Better Autogen Coaching Agents

Building an AI agent that knows when to talk is one thing, but how do you make sure it’s saying the right thing?

In my earlier geek-out time, I used AutoGen to orchestrate a career-coaching chatbot ( https://medium.com/p/37ddbcc38d2e ). It did a good job of routing users to the right expert, but there was a significant gap: quality control. The agents would sometimes give direct advice instead of asking open, reflective questions, or they’d ramble instead of being concise. It became clear that simply coordinating agents wasn’t enough; I needed to guarantee a consistently high-quality, aligned output from each one.

This is a common challenge in AI, but it’s where DSPy ( https://dspy.ai/ ) comes in. While it excels at ensuring quality outputs — as I’ll show you — its core philosophy goes far beyond just that. DSPy treats LLM like programmable components, shifting the focus from the brittle art of prompt engineering to the systematic engineering of reliable AI applications.

This is an excerpt — the full article continues on Medium.