PUBLIC SAFETY RAPID DECISION SUPPORT SYSTEM · ALAYA PUBLIC SAFETY LAB
Which vantage point sees which exit, whether the cordon's radios reach, where a person on the run goes: the computation is the same set of engines as in the Tactical Lab and the Rescue Lab, rerun here in police terms. The upper layer is ZERO turning a request into a computable question; the lower layer is the tools it calls. Defence is the core of the system; police work is the dual-use extension of the same stack, with the model and the data kept on each agency's own premises.
The language model, deployable on-prem, does one thing: it turns a sentence into a computable question and lays the answer out with its sources.
Every card is a page that really computes; the ones marked “by arrangement” run on ZERO's own host.
Which vantage point sees which exit and where the blind spots are: observation points and drone altitude.
Reach of the cordon, relays and units on scene, and which block stays silent.
Where to fly first with limited sorties: the search console for missing persons and fleeing suspects.
The ordering engine traffic control needs after a disaster: which cut first, when each village gets its road back.
What the aircraft do when the link drops: the swarm-autonomy demonstration.
Where a person missing in the water drifts: the drift and survival weighting inside the sea mission.
Start from one alert and watch ZERO pick its tools, propose and wait for approval (base-defence alert demo).
Each scenario lists the tools ZERO would call; the only numbers quoted are the engine's published simulation comparison, never results from a police scenario.
Three drones working together on real streets, the building's surroundings and exits first; the line-of-sight engine works out which vantage point sees which exit and where the blind spots are.
Currently the Hsinchu city search screen (real streets, three drones); adaptable.
Last known position, time, speed on foot or by car: probability × value search decides where limited sorties fly first; one new clue recomputes the whole probability map and re-tasks the fleet.
| ALAYA weighted search | Conventional | |
|---|---|---|
| People found (same engine, wildfire scenario) | 93% | 65% |
Numbers: the same engine's 200-run synthetic comparison on the wildfire survivor scenario (assuming the intelligence is correct); they are not results from a police scenario.
Layered perimeter watch and a human-approved response ladder: detect on the outer layer, confirm on the middle, respond on the inner, with the approver named at every layer.
Currently the facility-defence version of the screen; adaptable.
Three-drone layered guidance and dynamic role reassignment is the technical base: reconnaissance, cordon, lighting, announcements, non-lethal drops, every drop approved by a person; no lethal autonomy.
Within the Police Weapons Use Act and the Civil Aviation Act's remotely piloted aircraft rules (per each agency's regulations); police drone procurement and unit status remain to be verified and are not stated here.
Bring a real situation; we rerun the same engines on your data and watch, on the spot, which tools it calls and what it proposes.