Real-time Multimodal AI: Translating Sights and Sounds into Action
The Challenge: Business data isn't confined to neat spreadsheets anymore — it's live video, audio, and physical scenes. The challenge is processing these fast-moving inputs in real time to make split-second decisions.
A fast-paced, hands-on building competition that brings technical capabilities to life. In under 100 seconds, a Gemini model acts as a sophisticated real-time commentator and objective judge — analyzing physical toy blocks for color, structure, and intricate detail. This demo serves as a powerful proof of concept, illustrating how rapidly AI can synthesize complex visual scenes into instant situational insights and meaningful feedback for participants.


A split-second track assistant for driving simulators. While the driver is on the track, the agent processes dashboard telemetry and track visual landmarks simultaneously, delivering instant voice coaching directly to the driver's headset.
Instant visual auditing: Turn complex video and photo streams into structured compliance reports, compressing multi-week operational review cycles down to minutes.
Personalized advisory at scale: Deliver tailored, context-aware coaching in real time — whether you're training employees on a high-speed assembly line or supporting students.
Who is Building with Gemini Today?
Kingfisher PLC: Developed the "Hello B&Q" assistant to help customers identify plumbing defects. Customers snap a photo of a broken part, and the assistant analyzes the image, cross-references it with their massive inventory, and finds the compatible replacement stock instantly.
Lush: Developed a multimodal checkout system ("Lush Lens") that automatically identifies packaging-free ("naked") bath and body products. By replacing manual scanning, it slashed peak holiday shopping queues from out-the-door to under three minutes, while saving 440,000 liters of water.
Jo Malone London & The Estée Lauder Companies: Co-developed an AI Scent Advisor. Customers describe a mood or memory in natural language, and the model maps their preferences directly to a complex olfactory database to deliver bespoke fragrance recommendations.