Building an AI Assistant with DSPy | by Lak Lakshmanan | Mar, 2024

A way to program and tune prompt-agnostic LLM agent pipelines

I hate prompt engineering. For one thing, I do not want to prostrate before a LLM (“you are the world’s best copywriter … “), bribe it (“I will tip you $10 if you …”), or nag it (“Make sure to …”). For another, prompts are brittle — small changes to prompts can cause major changes to the output. This makes it hard to develop repeatable functionality using LLMs.

Unfortunately, developing LLM-based applications…