Icebreaker AI

Inspiration

I attend a lot of professional events. Manually researching speakers and judges before each event takes a significant amount of time and mental energy. I want to focus on showing up prepared, confident, and energized, instead of spending hours doing pre-event research.

What it does

The system scans my Google Calendar for upcoming events, extracts the associated Luma event links, and parses the event pages to identify speakers and judges. For each person, it looks up their company information and most recent LinkedIn activity, combines that context with my CV, and generates personalized talking points and ice-breaking questions. All talking points are compiled and emailed to me ahead of the event.

How we built it

I built the entire workflow using the Retool platform, using agentic workflows to orchestrate calendar ingestion, event parsing, external research, and talking-point generation.

Challenges we ran into

The agent was initially non-deterministic. To fix this, I broke the logic into smaller, modular agents and used Retool Workflows to enforce execution order and consistency. For example, without explicit iteration control, the agent would sometimes skip events in my calendar.

Accomplishments we're proud of

Researching one person manually takes around 10 minutes. A typical event has about 30 people, and I attend at least two events per week. This system saves me roughly 10 hours of research every week, allowing me to focus entirely on showing up and driving meaningful conversations.

What we learned

Providing additional context to agents as plain text (.txt) files is significantly faster and more reliable than using PDFs.

What's next

Add voice output so I can listen to talking points while commuting, making it easier to prepare on the go without reading or dealing with motion sickness.

Built With

• retool