← All case studies

An AI PC agent that learns by demonstration

Workbeaver's AI agent automates tasks on a PC after the user shows it how to do them once. We worked across the desktop client and the cloud services behind it.

Delivered by our founder as a consultant to Workbeaver.

ClientWorkbeaver
DisciplineSoftware development
Our roleFull stack development, consulting
PeriodAug 2024 to Apr 2026
2Clouds: GCP and AWS
20 moEngagement length
01 / The challenge

Turning a human demonstration into a repeatable automation means capturing what happened on screen, reasoning about it with large language models, and replaying it reliably on a different machine. That needs a responsive desktop client and backend services that scale on more than one cloud.

02 / What we built
  • Microservices on Google Cloud and AWS for the agent's backend.
  • Server-side logic in Python with FastAPI.
  • Desktop and web client work in TypeScript with Next.js and Electron.
  • Integration with Gemini and OpenAI models.
03 / Stack
TypeScriptNext.jsElectronPythonFastAPIGeminiOpenAIGCPAWS
Next project

Have a system to build or fix?

Software, electronics, or both. Your message goes straight to the architect, not a sales team.

Start a project