RAISE Summit Hackathon.
Mantra explores how AI agents can test apps, compare variants, and help decide what to build next.
3rd place Google DeepMind track
One continuous testing loop.
Test
Run parallel browser sessions through app interactions
Vary
Generate builds around explicit design hypotheses
Create
Produce ad-video concepts with Veo and Nemotron
Decide
Compare variants and recommend keep, iterate, or stop
Building the demo and its AI tools.
I worked on the Three.js game used by the agents, including its environment, animated characters, responsive interface, and rendering performance. I also built a Gemini-based video-tagging library with structured outputs and contract tests, giving the team a reusable way to describe recorded footage.
Building Mantra together at the RAISE Summit Hackathon.
3rd.
Google DeepMind track / RAISE Summit Hackathon
The team placed third in the Google DeepMind track. The demo connects browser playtesting, generated variants, and a decision dashboard. Its advertising integration creates paused test campaigns and uses seeded metrics; it does not demonstrate live campaign performance or paid acquisition results.
Certificate of Excellence / 3rd place, Google DeepMind track / RAISE Summit Hackathon, July 2026
With the Google DeepMind track winners at the RAISE Summit Hackathon.


