Inspiration

On August 2, 2026, a propane grill on a first-floor balcony at 815 Sea Spray Lane in Foster City started a fire that grew from one unit to a six-alarm blaze in under two hours. Roughly 50 units burned, more than 150 people lost their homes, four firefighters were injured, and it took over 12 hours to put out. Dispatch received call after call from residents saying they were trapped. Fire officials later said the building's layout itself made containment hard.

That's one story out of many. There are around 1.4 million fires annually, causing over 3,900 deaths.

That gap is what Sentinel is built to close.

What it does

Sentinel is a fire emergency platform with two sides, a resident side and a firefighter side, built on one shared live model of the building.

1. Model the scene. Sentinel fuses firefighter bodycams, building cameras, residents' phones, and floor plans into one live 3D model. Every resident and responder is a position on the model. Fire, smoke, and blocked corridors are marked as they spread. New observations refine the model even when the original floor plan is incomplete or wrong.

2. Route everyone. Each resident is routed to the nearest safe exit, or told to shelter in place if the hallway is already unsafe. Firefighters are routed by urgency, not distance: a crew will pass a stable resident on floor 1 to reach a critical one on floor 2, while a second crew picks up the first. When a corridor becomes blocked, every affected route replans and both sides see the change.

3. See the path. The same route is shown two ways. Command watches the live 3D model. The person in the building, resident or firefighter, follows AR arrows on their phone along that route, with the next turn and distance in view and a warning when an exit ahead is blocked.

4. Talk to them. Residents reach responders by chat or voice. Every message arrives pinned to the resident's coordinates on the model, so crews know who is where and what they need. "Smoke in the hallway, I can't reach the stairs" shows up on the commander's screen next to a dot on floor 2, unit 204, and the reply "Stay inside, close the door, F1 is 40 m out via Stair B" goes back to that person.

Why this fits "Build something AI cannot do alone"

Sentinel's routing and priority logic is only as good as what it knows, and in a fire almost everything it needs to know comes from people.

The model is built from human eyes: the firefighter wearing the bodycam, the resident who reports smoke, the commander who marks a collapsed stairwell. AI fuses the inputs; humans supply them.
The AI proposes routes and priorities. Firefighters confirm or override them. A commander outside sees things the model cannot, like a wall that looks ready to fall, and the model updates from that call.
Residents are not passive dots. Their messages change the plan. A single "the door is jammed" reroutes a crew.
The AR view and the chat exist so that the AI's output lands with a person who can act on it, and so that the person's reply flows back into the model.

Alone, an AI with a stale floor plan would route people into smoke. Alone, humans in a burning building cannot see each other. Together, with one shared model in the loop, each one covers what the other cannot.

What's next

Real sensor inputs: smoke detector events per unit, phone check-ins, and bodycam position tracking.
Pre-incident plans so Sentinel starts with a known graph for buildings that have one.
Running the model on a laptop in the command vehicle so it works when the building's network is gone.
Working with a local fire department to test the firefighter view in a live-burn training facility.
Demo link is in the "Try it out" links