Hands-on Workshop: Build Real-Time AI Agents (Pune Builder Session)
Most AI agents are still blind. They can reason, plan, and generate — but they can't see your screen, hear your meeting, or act on what's happening around them in real time.
This is the perception bottleneck. And we're going to break it open together.
Join us for a hands-on builder session where we ship real-time AI workflows from scratch — no slides, no theory, just working demos and code you can take home.
What we'll build and ship:
1. Claude Code × VideoDB Skills Watch Claude Code reason over live video context using VideoDB Skills — structured, reusable building blocks that let agents operate on real-time streams with a single prompt. We'll show how to compose agent workflows that understand what's happening on screen, not just what you describe to them.
2. Real-time AI Workflows See the full See → Ingest → Act loop in action. From a live desktop capture, we'll run transcript generation, semantic audio indexing, and visual scene understanding — all under 2 seconds of latency. Practical patterns for meeting copilots, in-call assistance, and screen-aware agents.
3. Desktop Perception with OpenClaw — Monitoring & Alerting We'll demo how to capture screen and system audio, attach AI-powered alert conditions in plain English, and trigger webhooks when those conditions fire — no custom models, just prompts. Think: "alert me when a pricing objection is raised" or "flag when sensitive content appears on screen."
4. DeepSearch over Live & Recorded Streams Once your stream is indexed, you can search it like a database. We'll show semantic search across indexed audio and visual content — jump to any moment, query decisions made in a meeting, or find exactly when something happened on screen.
🛠 Come ready to build: Bring your laptop. We'll have starter code ready. You'll leave with working pipelines.
Who this is for: AI/ML engineers, full-stack developers, agent builders, and anyone building tools that need to understand the real world — not just text.