From Raw Calls to a Clickable MVP in One Afternoon

Every AI-for-PM listicle this year says the same thing: drop your interview transcripts into an LLM and get themes in minutes. That part is true, and it's also where most PMs stop.

The interesting work happens between "themes" and "a v1 someone can click":

• Fixing transcripts that misattribute who said what
• Finding where two stakeholders want opposite things
• Writing a cut list with reasons
• Choosing the one piece worth prototyping

In this session I'll walk through the full pipeline I used on a real two-persona discovery project, rebuilt here as a fictional case. You leave with every prompt.

What you will walk away with

• A 7-step prompt chain that takes two raw interview transcripts to a scoped v1 and a prototype brief, ready to reuse on your next discovery round
• A way to pick what to prototype: rank the pieces of your v1 by risk instead of picking the most fun screen
• The "stated vs. revealed" lens: spot where the personas collide, and the signals a generic AI summary drops

Who it is for

PMs who already use Claude or ChatGPT on research data and have vibe-coded at least once, but haven't chained discovery, scoping and prototyping into one repeatable pipeline.