Turning narrated demos into scalable agentic knowledge
Instead of asking a team to watch recordings they will forget tomorrow, I turned two narrated demos into knowledge an agent serves at the moment of need.
Everything tagged ai-experiments, newest first.
Instead of asking a team to watch recordings they will forget tomorrow, I turned two narrated demos into knowledge an agent serves at the moment of need.
A custom web client that swaps the ADK camera feed for screen capture, so the agent watches the application instead of the user.
Arguing over translation versus localisation misses the point; thinking in layers of context maps cleanly onto a multi-agent architecture.
Instead of buying an expensive translation tool, I built an orchestrated localisation agent in Copilot Studio that never leaves the Microsoft 365 tenant.
Getting Google's ADK bidirectional streaming demo running locally with mic, camera, and voice, as the foundation for a live in-app assistant.
While OpenAI and Anthropic fought over a Pentagon deal, Google shipped WebMCP, UCP, and A2A, the protocol stack for how agents will use the web.
Working notes from recreating a voice-and-screen assistant: one path failed on authentication, the next produced a working streaming prototype.