Real world
A person enters a fitting room carrying garments. This is the scene — illustrated, never recorded.
IllustrationNo capture
Nextop platform · Physical AI
Spaces can understand what happens inside — without watching people.
Low-cost, self-configuring Spatial AI for fitting rooms, restrooms, hospitality, healthcare environments and industry.
No video leaves the room — because no video is created.
No RGB video. No face recognition. Anonymous event-level analytics by design.
What a camera seesImages · identity · privacy burden
What Spatial AI understandsSignals · anonymous events · no images
Drag to compare
Live event stream
Anonymous, event-level output
- TrackPerson #A1
- Entered14:32:08
See the difference
One scenario. Three ways to read it.
A person enters a fitting room carrying garments. Follow the same moment from physical world to spatial intelligence.
Raw multimodal sensing
mmWave point cloud and range-Doppler-style returns, ultrasound echo arcs, and optional Wi-Fi CSI waveform. Physical signal, no pixels.
mmWaveUltrasoundWi-Fi CSI (optional)
- 01Entry
- 02Occupancy 1
- 03Object interaction
- 04Activity pattern
- 05Exit
- 06Room state Δ detected
AI output
Local inference turns those signals into an anonymous event sequence — the only thing the system keeps.
AnonymousEvent-level
Physical worldSignals, not imagesSpatial intelligence
Why now / the gap
Rich spatial data has always cost you a camera.
There is a gap between sensors that see too much and sensors that barely see anything at all.
Camera
Rich data, heavy burden
- Detailed visual data
- Image capture of people
- Privacy and consent burden
- Often unacceptable in private spaces
Traditional occupancy sensor
Presence only
- Occupied / free
- No activity context
- No physical change detection
- No spatial understanding
Nextop Spatial AI
Anonymous spatial understanding
- Anonymous occupancy and flow
- Activity patterns (prototype targets)
- Physical change vs baseline
- Continuously learning spatial state
How it works
Four steps, one compact node.
A single node combines 60 GHz mmWave radar, ultrasound/acoustic sensing and optional Wi-Fi CSI / ToF. Nothing it produces is an image.
Sense
60 GHz mmWave radar + ultrasound/acoustic sensing, with optional Wi-Fi CSI and ToF modules for extra spatial detail.
Understand
Local inference on ESP32-class edge hardware turns raw physical signals into anonymous features and events — in the room, in real time.
Learn
The node learns a baseline of the room: its normal geometry, normal rhythms, normal occupancy — so it can detect what changed.
Sync
Periodic sync to a more powerful edge/cloud model updates the spatial model — a digital twin of the environment's physical state.
Designed so rapid inference stays local and only anonymous event-level data leaves the room.
One node or many
Start with one. Add nodes only when the space asks for it.
Nodes auto-discover each other, self-calibrate and fuse their observations. Coverage grows without a redesign.
Small room
1 nodeFitting room, restroom cubicle, single office. One node covers the volume.
Medium environment
2 nodesHotel room, clinic room, small retail zone. Two nodes reduce blind spots.
Complex environment
4 nodesOpen floor, industrial cell, multi-zone area. Four nodes handle occlusion and long spans.
Mount. Power. Connect. Calibrate.
Add nodes only when coverage requires it.
- 01Mount
- 02Power
- 03Connect
- 04Calibrate
Simple installation
Power and Wi-Fi, mount on wall or ceiling, calibrate, go.
Self-configuration
Nodes map their own view of the room and set zone boundaries automatically.
Automatic fusion
Designed so multiple nodes merge observations into one shared spatial state.
Add only when needed
More nodes for more coverage or harder occlusion — never required by default.
Levels
Three levels of spatial understanding.
Level 1 is the working foundation. Levels 2 and 3 are prototype targets — capabilities the platform is designed to deliver as models are validated.
Level 1Foundation
Presence & Flow
- Occupied / free
- Entry and exit events
- Dwell time
- Anonymous movement paths
Level 2Prototype target
Activity Intelligence
- Walking, sitting, bending
- Object interaction
- Prolonged inactivity
- Zone-level behaviour
Level 3Prototype target
Spatial Twin
- Baseline room state
- Change detection
- Before / after comparison
- Periodically refreshed digital twin
Physical Δ
The room after is not the room before.
The node learns a baseline, then reports the difference — without ever forming an image of the person who caused it.
Baseline
After exit
Δ
Drag the handle
Baseline
- Bench clear
- 3 garments on hooks
- No bag present
After exit
- Bench occupied
- 2 garments on hooks
- Bag left behind · Zone B
Output
Δ Zone B- 01Person exited
- 02Occupancy 0
- 03Spatial state differs from baseline
- 04New physical change detected in Zone B
Physical change detection is a prototype-target capability the system is designed to deliver.
Use cases
Intelligence for spaces where cameras don't belong.
Every scenario below is served by the same node and the same anonymous event model.
Node 01Fitting rooms
Occupancy, dwell time and interaction patterns — plus possible item-left-behind and room-state change detection. No cameras while people change clothes.
OccupancyDwellRoom-state Δ
Node 01Smart restrooms
Occupancy, peak times, dwell, utilization and activity patterns. Can enable cleaning triggers and anomaly detection — without video.
UtilizationCleaning triggersAnomalies
Node 01Hotels & hospitality
Anonymous occupancy, room-state change between sessions, and operational insight for housekeeping. Optional multi-node coverage for suites.
TurnoverRoom-state ΔMulti-node
Node 01Healthcare & assisted environments
Movement, bed and room activity, and prolonged inactivity signals. Fall-like-event detection is a research direction — a future, validation-dependent capability, not a medical device or diagnosis.
ActivityInactivityResearch direction
Node 01Industry & restricted areas
Anonymous zone activity, procedure sequence as Proof of Physical Work, and obstacle or physical-state change in controlled areas.
Zone activityProof of Physical WorkObstacle Δ
Node 01Retail & private spaces
Zone-level behaviour analytics designed for places where a camera would be undesirable or simply not allowed.
Zone flowDwellNo video
The node
One compact node. No GPU in the room.
Conceptual layout — not final industrial design.
- 01
ESP32-class edge MCU
Local feature extraction and inference
- 02
60 GHz mmWave radar
Range, motion and micro-movement returns
- 03
Ultrasound TX / RX
Acoustic echo for geometry and object change
- 04
Optional Wi-Fi CSI / ToF
Modular add-ons for extra spatial detail
- 05
Power + Wi-Fi
Single supply, standard wireless connection
No dedicated GPU per room. No camera module. No image sensor of any kind.
60 GHz mmWave radarUltrasound TX / RXEdge AI on the nodeOptional Wi-Fi CSI / ToF
Simple to deploy
Mount. Power. Connect. Self-calibrate.
A compact node goes up high in the room like a detector — no camera, no wiring project, no GPU in the space.
- 01
Step 01
Mount
Wall or ceiling, high in the room, in minutes.
- 02
Step 02
Power
One low-voltage supply. No dedicated cabling project.
- 03
Step 03
Connect
Standard Wi-Fi. The node joins the site and appears in the platform.
- 04
Step 04
Self-calibrate
The node maps its own view, sets zone boundaries and learns the room baseline.
Add another node when you need more coverage.
Auto-fusionPrivacy by architecture
It cannot leak an image it never captured.
Privacy here is a consequence of the sensing stack, not a setting. This describes the architecture — it is not a claim of guaranteed regulatory compliance.
RGB video
No RGB video · No face recognition
Anonymous features & events
Features · Events
- No RGB video
- No face recognition
- No raw voice storage
- Anonymous ephemeral tracks
- Local feature extraction
- Event-level cloud sync
- 01Physical wavesmmWave · ultrasound · optional CSI/ToF
- 02FeaturesExtracted locally on the node
- 03EventsAnonymous, event-level records
Never images.
No RGB video. No face recognition. Anonymous event-level analytics by design.
The innovation
The hardware is the cheap part. The model is the product.
The long-term innovation is a continuously learning spatial model plus multimodal sensor fusion — not the node alone.
Multimodal sensing fusion
Radar, ultrasound and optional Wi-Fi CSI / ToF are fused into one coherent read of a space — each modality covering the others' blind spots.
Physical Δ
The system learns a baseline and reports what changed against it: geometry, objects, room state — the difference, not the picture.
Distributed spatial intelligence
Cheap nodes handle fast local inference; a periodic deeper model refreshes the spatial model and digital twin of the environment.
Long-term moat
A multimodal spatial dataset built from real environments, and spatial models that become reusable across new rooms, buildings and verticals.
From one space to an intelligent building
One room today. A whole site as you grow.
Every node learns locally. The platform is designed to merge those observations into one unified, anonymous spatial view of the site.
- Stage 011 node · 1 room
One node, one room
A single node covers a fitting room, restroom cubicle or small office: occupancy, anonymous flow, dwell and physical change against a learned baseline.
- Stage 022–4 nodes · complex room
Multiple nodes, one complex space
Add nodes for long spans, occlusion and multi-zone layouts. They auto-discover each other and are designed to fuse into a single room state.
- Stage 03Many rooms · one floor
Multiple rooms, one floor
Rooms become a floor: comparable metrics per space, anonymous movement between zones, and operational patterns across the whole area.
- Stage 04Site-wide spatial model
Site-wide spatial intelligence
A periodically refreshed spatial model — a digital twin of physical state — designed to span the entire site, still without images or identities.
Anonymous, event-level data only. Reusing learned spatial models across new spaces is a design goal as deployments grow.
Bring intelligence to spaces where cameras don't belong.
We are selecting pilot partners for fitting rooms, restrooms, hospitality, care environments and industrial zones.
www.nextop.cloud