Building Collaborative Prompt Engineering Playgrounds Using Jupyter Notebook

LinkedIn’s engineering team built shared Jupyter notebooks for testing prompts so sales experts could evaluate AI features using the same underlying workflows as engineers. The notebooks load the application’s code, prompt templates and test data, letting participants change an instruction and run the resulting workflow interactively. That setup matters because a useful prompt must work with the product’s actual data gathering and sequence of model calls. An isolated chat test can miss those interactions. By keeping notebooks and datasets in the code repository, the team also put changes through review. The company describes a way to bring domain expertise into product development while retaining a common, reviewable implementation.