(Deep) Dive Copilot

Dive Copilot explores how voice AI can become an operational interface for professional underwater work. Divers often need to inspect structures, record findings, follow mission tasks and communicate with surface control while their hands and visual attention are already occupied. Important information is frequently communicated verbally during the dive, but that information may later need to be manually reconstructed, transcribed or entered into another system. Dive Copilot addresses this by connecting natural speech directly to structured mission state. A diver can say: "Record moderate corrosion at support joint C2." and the system creates a structured observation. They can say: "I've reached C2." and the system updates confirmed position. They can correct previous information: "Actually, make that slight corrosion.", complete a task: "Inspection of C2 is complete.", or communicate with surface control: "Ask surface if they want a closer inspection." The prototype is built around one central idea: Voice should be an interface to operational state, not merely a conversational layer. The current implementation combines the AssemblyAI Voice Agent API, mission-specific speech context, structured HTTP tools, a FastAPI mission backend and a browser-based simulated diver HUD / surface-control interface. The prototype currently supports three mission scenarios: - Underwater Infrastructure Inspection; - Reef Biodiversity Survey; - Marine Debris Survey. The same configurable agent adapts to each mission by changing its speech-recognition context, terminology and mission workflow.