Groundwork AI
From a business idea to validated customer insight. It researches the market, finds who to talk to, runs voice interviews with an AI researcher and rewrites its own questions as it learns.
- Role
- TODO: Martin's role
- Team
- TODO: team
- When
- GTM hackathon sprint with Lovable and JPMorgan Chase · Nov 2026
The problem
Talking to customers is the most useful thing a founder can do and the first thing that gets skipped. Finding the right people, booking calls and asking good questions takes weeks.
What we built
Groundwork AI takes an idea and a website. It researches competitors and drafts an ideal customer profile, then reaches out to candidates and runs short voice interviews with an AI researcher. At the end it writes a report on what people actually said.
Decisions I’m proud of
The questions get better on their own. After a batch of interviews the tool looks for patterns, like 8 of 12 people bringing up the same pain, and adds a question about it to the next version of the guide. That adaptive guide was the part we refused to fake.
Interview like a researcher, not a salesperson. The voice agent asks about real past events instead of hypotheticals, asks one follow-up when an answer is vague, and never pitches a product.
Be honest about the demo. We wrote down what was real and what was seeded before we built anything. The research, the voice interviews and the adaptive guide were real. Candidates came from an uploaded list instead of scraping, and SMS only went to teammates who had opted in.
Where it is now
We built it in a GTM hackathon sprint with Lovable and JPMorgan Chase. The demo is live at groundworkai.lovable.app.

A live interview. The AI asks about a real past event and follows up on the answer

The guide learns as it calls. Each version adds a question based on what earlier interviews found

Research, AI interviews and insight, the three steps from idea to evidence