Press Cmd/Ctrl + P to export: choose Landscape, set Margins to None, and enable Background graphics to preserve the full visual design.
Back to Deck
PhysenseAI
Physical Space Intelligence — enabling AI to understand what is happening inside a space
Investor Deck · 2026 · PhysenseAI
Every Data Modality Creates Its Own Intelligence Layer
Text has LLMs. Images have Vision AI. The physical world still lacks an intelligence layer.
We are building the intelligence layer for physical space.
Devices Report Signals. They Do Not Reconstruct Events.
Sensor events ≠ real-world events. The missing step is understanding.
- 02:10 · Bed pressure sensor → No pressure
- 02:11 · Bathroom motion sensor → Occupied
- 02:11 · Door contact → Open
- 04:10 · Bathroom motion sensor → Still occupied
Discrete, unconnected and memoryless — able to trigger only isolated rule-based alerts
≠
Real-world Event
- 01Resident leaves bed at night
- 02Enters the bathroom
- 03Remains in the bathroom for two hours
- 04A serious incident may have occurred
Continuous, contextual and time-aware — enabling meaningful risk assessment
What is missing is not more sensors, but the layer that turns signals into events.
Three Curves Are Converging Now
Mature sensors, accessible compute and rising Physical AI demand make this layer possible now.
Sensors Are Now Abundant and Affordable
Millimeter-wave radar, infrared, door contacts, pressure and environmental sensors are mass-produced and affordable enough for room-level deployment. Spaces no longer lack signals.
Edge Compute Has Become a Commodity
Edge NPUs make continuous local inference practical. Raw data no longer needs to leave the premises, unlocking privacy-sensitive environments.
AI Is Moving Into the Physical World
Robots, embodied AI and intelligent buildings all need shared spatial context to act. This layer has yet to become infrastructure.
The window is open: signals are abundant, but understanding is still missing.
We Build the Layer Between Devices and Applications
We build the intelligence layer every spatial application needs — not sensors, not a single application.
Care · Robotics · Buildings · Security
Event reconstruction · Spatial state · Behavioral baseline
Gateways · Protocols · Time sync
Radar · Infrared · Door contact · Environment
| IoT Platform | PhysenseAI | |
|---|---|---|
| Data form | Individual device readings | Continuously updated spatial state |
| Time dimension | A value at one moment | Continuous events and history |
| Decision logic | Threshold-based alerts | Comparison with historical baseline |
| Output | Raw data / reports | Events, states and anticipation |
Build the layer once, reuse it across every spatial application.
An Eight-Step Reasoning Chain: From Signals to Action
The intelligence lives in the model and reasoning engine — independent of any hardware stack.
- 01
Multi-source Input
Any combination of sensors
- 02
Signal Processing
Denoising and timestamp alignment
- 03
Spatial Representation
Entities + space + time
- 04
Event Reconstruction
Discrete signals → continuous events
- 05
Spatial State
Who / where / doing what
- 06
Behavioral Trace
Activity changes over time
- 07
Deviation Detection
Compared with personal history
- 08
Anticipation & Guidance
Risk level + actionable response
The output is not a data point, but a continuously updated state of the space.
A Reasoning Engine Today.
A Learned Spatial Model Tomorrow.
We do not call a rules engine a foundation model. This is where we are today — and where we are going.
Spatial Reasoning Engine
- Multi-source fusion and event reconstruction
- Rules + statistical models detect deviations
- Explainable and auditable outputs
- Light adaptation for each space
Behavioral Baseline Learning
- Automatically learn a baseline for each person and space
- Move from manual to self-learning thresholds
- Recognize patterns across rooms and day/night cycles
- Now validating with real device data
Learned Spatial World Model
- Train on large-scale real-space data
- Generalize directly to unseen spaces
- Predict the next state, not just explain the current one
- Become shared context for Physical AI
The path is clear: use the engine to access real spaces, then use real spaces to train the model.
The Real Value: Comparing Each Person With Their Own Baseline
Not one threshold for everyone, but each person compared with their own historical norm.
Event
One bed exit, one bathroom entry, one door opening
Event Sequence
Bed exit → bathroom → return to bed, with duration for each stage
Behavioral Baseline
How often this person rises at night, for how long, and how far
Deviation & Anticipation
Tonight deviates sharply from personal norms, triggering a risk level and recommended action
Note: The system outputs spatial states and risk signals to support alerts and response workflows. It is not used for medical diagnosis.
The same event can be normal for one person and a risk signal for another.
What Is Proven — and What We Are Validating
Everything below is verified on real devices. Nothing is claimed before it happens.
- Real-device ingestion verified: multiple sensor types feed one pipeline and produce stable event streams after timestamp alignment
- Event reconstruction works: discrete signals combine into ordered spatial event sequences
- Spatial state streams continuously: in bed, out of bed, bathroom occupied and vacant states update over time
- Edge execution verified: inference runs locally without uploading raw data
- Self-learning personal baselines to replace manually configured thresholds
- Stability and false-positive rates across multiple rooms and floors
- Continuous overnight operation in real-world settings
- On-site refinement of the live care-station dashboard and response loop
Simulated data validates reasoning logic only; it does not represent field performance.
The engine works. This round is about continuous operation data from real spaces.
One Core, Open at Both Ends
Swap sensors on the input side or applications on the output side without changing the core.
Any device; protocol and time alignment
Entity · Space · Time · State
Events · Baseline · Deviation · Anticipation
Events, states, risk levels and recommended actions
Hardware and applications will change. The spatial intelligence layer compounds over time.
Start With One High-Value Space: Senior Care Facilities
Not because care is our only market, but because it proves the value of spatial intelligence fastest.
High Value
Nighttime falls and bathroom overstays carry severe human and liability costs
High Frequency
The workflow runs every day and night, making value continuously visible
Privacy Critical
Cameras are unsuitable in bedrooms and bathrooms; non-visual sensing is essential
Context Dependent
Whether a signal indicates risk depends entirely on the person’s normal behavior
Capabilities proven here transfer directly to other spaces.
AI Spatial Care System
Customers are not buying sensors. They are buying a clear, live understanding of what is happening across the building.
- 3F · 12 Rooms
- 2F · 12 Rooms
- 1F · 10 Rooms
- 3F · 12 Rooms
- 302Bathroom overstay 2h00m
- 304Out of bed 6m at night
- 307Bathroom occupied, normal range
Night bathroom stay > 15m → Watch; > 30m → Alert
Notification path: Staff dashboard → Mobile push → SMS escalation after 5m without response · Sense → Understand → Alert → Respond
Who Buys, Who Signs, and How We Enter
Prove the overnight response loop on one floor, then expand by building and portfolio.
Target Customers
- Mid-to-large senior care facilities
- Integrated care and rehabilitation centers
- Premium communities and home-care providers
Decision Makers
- Facility director / COO: liability and reputation
- Head of care: night staffing and response efficiency
- IT lead: integration with existing systems
Entry Motion
- Pilot one floor, priced per room
- Prove the overnight response loop and false-positive rate
- Expand building-wide, then across the operator’s portfolio
The buyer is clear. This round is about securing the first referenceable pilots.
Project → Product → Platform
Use projects to access real spaces, turn capabilities into a product, then build a platform.
Project Delivery
- Deploy by facility and price by room
- One-time deployment fee + annual service fee
- Gain access to real spaces and real data
Standard Product
- Standard hardware kit + software subscription
- Shorter deployments and higher gross margin
- Repeatable through channels and integrators
Capability Platform
- Spatial intelligence delivered via API
- Robotics, buildings and security pay per call
- Revenue decouples from deployment labor
Each stage compounds the data and reusable capabilities needed for the next.
One Core, Multiple Spaces
Scenarios and thresholds change. The core remains: events, states, baselines and deviations.
- Event Reconstruction
- Spatial State
- Behavioral Baseline
- Deviation & Anticipation
Care Spaces
Night bed exits and bathroom overstays · Fall-risk alerts and care-round dispatch · Long-term mobility changes
Buildings & Energy
Adjust HVAC and lighting to actual occupancy · Space utilization and workstation analytics · Reduce energy automatically in vacant areas
Security & Industrial
Loitering in restricted areas · Worker safety state in operating zones · Abnormal routes and timing
More spaces make the same core stronger — that is what defines a platform.
Two Flywheels Converge on One Outcome
The business flywheel wins spaces; the intelligence flywheel turns spaces into capability. Both drive More Real Spaces.
Business Engine
- 01Deploy one space
- 02Close the response loop
- 03Renew and expand
- 04Access more spaces
Intelligence Engine
- 01Real-space data
- 02Iterate baselines and models
- 03Fewer false alerts / better anticipation
- 04Faster delivery
Data use: customer data improves models only with explicit authorization and data minimization; raw signals remain on premises by default.
Spaces create the model. The model makes it easier to win more spaces.
Robots Need the Context They Cannot See
We do not build robots. We provide the environmental context they lack.
The Robot’s Own View
- Sees only a small area directly ahead
- Cannot see behind walls or into adjacent rooms
- Has no memory of the space
Environment-Level Spatial Context
- Live state of the entire floor
- Who is where and what just happened
- Which areas should not be entered now
Spatial context will become shared infrastructure for embodied intelligence.
Three Market Layers. Three Layers of Moat.
No speculative bottom-up TAM. Only directional anchors supported by public data.
Senior Care Spaces
Room-based care systems
Cross-Device Event Reconstruction
Align multi-vendor signals into usable event streams.
Buildings · Security · Industrial
One core, adapted by scenario
Continuous Behavioral Data From Real Spaces
Real-world data cannot be bought; it must be earned through deployment.
Physical AI Context
Used by robots and embodied AI
A Context Interface Applications and Robots Depend On
Once systems depend on the context layer, switching costs are substantial.
The moat is not one algorithm. It is the number of real spaces.
Team, Roadmap and This Round
This round has one clear goal: turn the engine into a system that runs continuously in real spaces.
Product definition, spatial intelligence solution and commercialization
Technical leader in spatial perception, heterogeneous sensor fusion and temporal intelligence; responsible for the core system from multi-source physical signals to event reconstruction, spatial state and behavioral reasoning, advancing it from prototype to real-world deployment and learned models
- NOW
Engine operational; real-device data ingestion verified
- 0–6 Months
First facility-floor pilots; overnight response loop live
- 6–12 Months
Self-learning baselines live; false-positive rate and response time become delivery metrics
- 12–18 Months
Standardize the product; validate a second vertical in buildings or security
- 40%R&D & Models
Reasoning engine, baseline learning and edge deployment
- 35%Deployment & Delivery
Pilot rollout, field refinement and delivery engineering
- 25%Team & Operations
Complete the core technical and delivery team
This round will deliver the first systems that run continuously in real spaces and can be formally accepted.
Give AI an Understanding of What Is Happening in Physical Space
We are not adding more sensors. We are building the capability for AI to understand physical space.