05 · GenAI proof of concept, 2023
CustomerGPT
A proof of concept for personalized sales-objection practice built when GPT-3.5-era models were much less reliable and before modern tool calling was widely available.
How it works
A learner selects a product and objection, then practices against an AI-generated customer. I used prompt structure, hidden conversational state, fixed interaction bounds, and application-side parsing to control how the simulation behaved rather than relying on the model to manage the experience on its own.
Outcome
The project was later presented to a Salesforce innovation team as an example of how generative AI could support personalized, just-in-time practice.
Tools
Articulate Storyline · JavaScript · OpenAI API