Peazy Trainer

Trainings in app
Template on Daytona

Inspiration

We wanted to move beyond passive “watch and learn” tutorials. Peazy Trainer was inspired by the idea of creating a safe, sandboxed space for interactive AI training, where users can learn by actually doing, with a voice coach guiding every step.

What it does

Peazy Trainer is a voice-led show-and-tell live coach app to learn new software in a safe environment. Gemini Live speaks the instructions, while BrowserUse Cloud performs the actions inside a sandboxed n8n workspace on Daytona. Each step is verified before moving on

How we built it

We built a React + Vite web app in a Training management app style, hosted via Cloud Run, that connects to the Gemini Live API for real-time audio and to BrowserUse Cloud for execution. The software to be learned(n8n in our case) was deployed with a pre made template as snapshot on Daytona. When the tutoring session starts, it automatically spins up a new Daytona sandbox and trains the user. Plans are defined in JSON .

Challenges we ran into

Starting Browser use actions on local device, Getting Browser use to follow preauthored commands instead of dynamic decisions, keeping Gemini’s speech and BrowserUse’s actions perfectly in sync, managing WebSocket session states, validating step outcomes were all surprisingly tough and ensuring all traces were being logged in galileo

Accomplishments that we're proud of

We got our first full speak-then-act loop running, Gemini spoke, verification passed, and the log displayed perfectly.

What we learned

Realtime Live API as brain is still very indeterministic with poor tool calls. Strict sequencing matters, deterministic promprint and structure beats spontaneity when coordinating AI actions. BrowserUse cannot kickoff in a real browser from a web app, needs native python support