Architecture overview

Automated Web Development Workflow

From a voice prompt in Gemini Live to a live web application on GitHub Pages, with no manual steps.

Executive summary

An end-to-end pipeline that connects voice-driven AI prompts, real-time cloud synchronization, agentic execution and automated web deployment.

How it works (video)

Architecture & data flow

  1. Gemini LiveUser prompt / file creation
  2. Google Drivefiles.watch push notification
  3. HTTPS webhookPush notification delivered
  4. IBM Bob HarnessGenerates code
  5. GitHub repositoryActions / Pages
  6. Live web applicationPublic URL

Detailed workflow

1. Voice interaction & concept capture

Gemini Live: primary multimodal interface.

  • User states website or app requirements verbally or in a live session.
  • Gemini Live formats structured specifications or source documents and saves them to Google Drive.

2. Event-driven webhook notification

Google Drive API: real-time event broker.

  • Drive monitors the target folder with files.watch / changes.watch.
  • On file creation or update it sends an HTTPS POST to the webhook endpoint.
  • Headers: X-Goog-Channel-ID (channel UUID), X-Goog-Resource-State (sync, add, update, content), X-Goog-Resource-URI (target resource).

3. Agentic execution & code generation

IBM Bob Harness: cloud execution environment and coding agent.

  • Validates webhook headers and responds immediately with HTTP 200 OK.
  • Uses a service-account OAuth 2.0 token to call files.get and fetch the updated document.
  • Parses the design brief and generates HTML, CSS, JavaScript or framework assets in a dedicated repository (e.g. web-builder-service).

4. Versioning & continuous deployment

GitHub & GitHub Pages: version control and hosting.

  • Repository isolation: the agent creates or updates a modular, dedicated repository.
  • Commit & push: generated code is committed via the GitHub REST API or SSH.
  • Automated build: a GitHub Actions workflow builds static assets.
  • Deployment: artifacts are published to GitHub Pages with an instant public URL.

Key advantages

Zero-latency orchestration
Push notifications trigger execution instantly, with no polling or manual file transfers.
Hands-free development
Natural-language voice prompts become fully deployed web applications end to end.
Modular architecture
Clean separation between interface (Gemini Live), storage (Google Drive), agent execution (IBM Bob Harness) and hosting (GitHub).
Traceability & versioning
A full audit trail across Google Drive documents and GitHub commit history.
Autonomous deployment
Dedicated repositories are provisioned and published to live public URLs via GitHub Pages.

System integration summary

This event-driven pipeline combines multimodal AI, serverless webhooks, agentic orchestration and CI/CD hosting into an autonomous web generation system.

Source code

The implementation lives in the IBM Bob Harness repository, on the feat/google-drive-gemini-live branch: github.com/BrUn3y/IBM_Bob_Harness/tree/feat/google-drive-gemini-live.