What you clicked
Traditional computing reads events: taps, swipes, the app you opened. It records what you did, never the state you were in while you did it.
Super Intelligent AI Systems for a Predictive World.
Predict the Future and Protect the Human™
One on-device loop: PAS intelligence, CascadeOS, and SmartSelfie hardware. It models state, ranks intent, and prepares the device before you ask.
- Equity round
- $50M
- Runway
- 36 mo
- Patents filed · 10 granted
- 13
- PAS architecture
- 7-layer
The shift
Computing is moving from reactive to predictive.
Existing devices respond after a user acts. A device that continuously reads physiological signal, behavioral history, and spatial context can instead model what the user is about to do, and prepare for it.
That capability does not sit in an app. It rests on a predictive computing layer embedded directly into the operating system and hardware stack, the substrate that turns a reactive tool into a continuously adaptive system.
Not a feature added to an existing phone. The phone, re-architected from the sensors up.
Traditional computing
- Command
- Response
Sentient computing
- Sense
- Understand
- Predict
- Prepare
The platform
One predictive platform. Three integrated layers.
SmartSelfie senses. PAS predicts. CascadeOS acts. One vertically integrated stack, designed together.
- 01The sensing device
SmartSelfie
The first physical product and continuous input layer. Where a conventional smartphone treats sensors as accessories to a screen, SmartSelfie treats the entire device perimeter as a sensing surface, fusing every signal on-device into a live behavioral state.
- 02The predictive intelligence
PAS(Predictive Adaptation System)
The Predictive Intelligence Infrastructure embedded into the OS and hardware stack. It models cognitive patterns, workflows, physiological tendencies, interface preferences, and how all of those evolve over time.
- 03The adaptive interface
CascadeOS
Turns PAS predictions into visible product behavior. PAS predicts, CascadeOS executes: suggest, preload, rearrange, or execute on the interface.
- SmartSelfie
- PAS
- CascadeOS
- Prepared action
How it works
From signal to action, on the device.
Four steps, running continuously on the device.
- 01L01–L02
Sense
SmartSelfie captures physiological, behavioral, environmental, and interaction signals continuously.
- 02L03
Understand
PAS fuses physiological, behavioral, and contextual inputs into a live estimate of context and user state.
- 03L04–L06
Predict
The system ranks likely next actions, sequences, and layouts on-device.
- 04L07
Prepare
CascadeOS suggests, preloads, rearranges, or executes, subject to confidence and risk policy.
Raw sensitive signals remain on-device whenever possible. A privacy-preserving cloud sits above as memory and refinement. The cloud trains. The device acts.
The seven cooperating layers behind these four steps
Reference document
PAS System Workflow Across Industries
Physiological Interface Bridge (PIB) in Action
From Ingestion to Understanding to Adaptation to Continuous Feedback
- Five stages
- Ingestion
- Physiological Interface Bridge (PIB)
- PAS Core Engine
- Adaptation & Action
- Feedback Loop
- Industries shown
- Healthcare
- Automotive
- Enterprise Software
- Mobile Gaming / Computing
- Robotics
View the diagramDownload PDFPDF · 1 page · letter
The intelligence
Most systems know what you clicked. PAS understands why, when, your cognitive state, and how your behavior evolves.
Predict the Future and Protect the Human™
PAS builds a behavioral digital twin of each user.
The Predictive Adaptation System is the Predictive Intelligence Infrastructure embedded into the OS and hardware stack.
A behavioral digital twin models a human: cognitive patterns, workflows, physiological tendencies, interface preferences, and how all of those evolve over time. That living model is what lets the system adapt to the person, not the average user.
The architecture
Seven cooperating layers, each optimized for its role.
PAS constructs the twin from continuously running layers: TinyML at the sensors, medium models for state estimation, larger models for prediction and orchestration, and a privacy-preserving cloud above as memory and refinement. The cloud trains. The device acts.
- L01Always-on sensingTinyMLTinyML on a low-power subsystem. Detects meaningful changes cheaply and instantly.
- L02Signal + feature extractionDSPRaw sensor streams become compact features: HRV, grip stability, conductivity, transitions.
- L03Fusion + state estimationMLPMedium models fuse physiological, behavioral, and contextual inputs into a live state.
- L04Prediction engineGRURanks likely next actions, sequences, and layouts. Tuned for on-device inference.
- L05Spatial interfaceGraphApp location and arrangement as behavioral signal. Layout is memory, not decoration.
- L06Language + intentSLMCompact on-device LM clarifies meaning, emotional tone, and ambiguous phrasing.
- L07Decision + orchestrationPolicyConfidence thresholds, safety rules, privacy policies. Chooses to act, suggest, or defer.
The sensing layer
The Physiological Interface Bridge.
The bridge is the mechanism through which devices obtain physiological understanding of the user, and it may be one of the most important inventions within PAS. Traditional computing knows what you clicked. PAS attempts to understand how you are: focused, stressed, tired, distracted, engaged, or overloaded.
How you are
The bridge attempts to understand the state behind the action: focused, stressed, tired, distracted, engaged, or overloaded. It turns physical contact into a sense of how the user is doing.
What makes the bridge unusual is that it changes form across devices. The same sensing-to-state layer is expressed through whatever surface the user is touching, then feeds physiology alongside behavior and context so PAS can make more accurate predictions.
One bridge, many forms
- Smartphone
- Contour sensor grid across edges, grip zones, and contact points.
- Smartwatch
- Wrist sensors in continuous contact with the skin.
- Vehicle
- Steering-wheel sensors, seat sensors, and cabin cameras.
- XR headset
- Head-contact sensors and eye tracking.
- Robot
- Interaction sensors, voice analysis, and behavioral observation.
On the SmartSelfie, the bridge takes the form of a Contour Sensor Grid integrated into the edges, grip zones, and ergonomic contact points. Its sensors may collect grip force, temperature, skin conductivity, motion, pulse-related signals, and respiration proxies.
The learning loop
The full system, in motion.
The device layers do not run once. They run in a loop with the cloud. The device senses and acts in real time; summaries flow up; the cloud accumulates memory, refines the model, and sends improved policies back. Every cycle makes the next prediction better.
On Device (edge)
- SensePPG, GSR, vision, grip, voice, and context, captured continuously.
- Edge feature extractionRaw signals become compact features on a low-power subsystem.
- Local PAS predictionOn-device models rank the likely next action, app, and layout.
- CascadeOS actionSuggest, preload, rearrange, or execute on the interface.
In Cloud
- IngestionReceives privacy-filtered event summaries, never raw sensor traces.
- Feature store + memoryTime-series behavioral history accumulates per user.
- Behavioral profile engineA persistent, evolving model of routines and preferences.
- Training + personalizationRetrains and ranks, then packages improved models.
The output / personal knowledge graph
The layers exist to build one thing.
Not a static profile. Not a recommendation engine. The behavioral digital twin is a personal knowledge graph: nodes for the apps, places, times, and physiological states a user moves through, and edges for the correlations PAS learns between them. Five signals compose it.
- Physiological Correlation
- Stress, focus, fatigue, excitement, and grip pressure fused from PPG, GSR, vision, and contour sensors.
- Temporal Learning
- Daily habits, work cycles, sleep and activity rhythms, communication timing.
- Contextual Intelligence
- Location, active task, device orientation, environment, social patterns.
- Interface Preference
- App layouts, window arrangements, workflow ordering, screen-state restoration.
- Continuous Reinforcement
- Accept and reject feedback, silent observation, longitudinal memory updates.
InputSignal description
Learned edges read like: Maps after leaving the office; rideshare when motion state rises; Calendar and Messages together before a meeting.
Oura measures. PAS acts.
Most physiological-sensing platforms stop at reporting. Sentient turns state into product behavior.
CascadeOS
The adaptive interface layer.
PAS predicts. CascadeOS executes. The intelligence lives in PAS; the value becomes visible on screen.
Once PAS has enough confidence, CascadeOS reshapes the workspace: stages apps and groups workflows around an active task, recalls the layout the user had before a meeting or at the same hour yesterday, and preloads the surfaces most likely to come next, ranked by PAS confidence.
Cascade Windows technology, US Patent 9,292,158, issued March 22, 2016.
Full IP schedule available in the data room. Request data room access.
The device
The phone is the first input.
The device perimeter is treated as a sensing surface rather than a frame around a screen. The industrial design is shaped by the sensing model itself, built around how a person actually holds, watches, moves, and speaks to a phone for hours a day.
The edge is the instrument.
Every module along the perimeter is a sensor. Nothing on the edge is trim. What a conventional phone spends on a frame, this device spends on capture.
Two finishes.
Blue
Red
Thin enough to disappear. Deep enough to sense.
The full sensing perimeter runs the length of the flank. The wave at the top and the notch at the bottom read in silhouette.
Six capture surfaces.
Signals are fused locally into a changing view of user state. Cloud processing is optional for real-time operation.
Rear camera cluster.
Front camera.
- Ambient microphone
- Short-form language and intent array. Not a transcription pipeline.
- Front camera
- Gaze and facial micro-signal capture for non-contact state estimation.
- Capacitive grip
- Contour-pressure sensors in the device perimeter read grip stability.
- PPG and GSR
- Physiological array in the outer perimeter edge streams heart rate, HRV, and skin conductivity.
- Inertial unit
- Motion and posture context. Commute, gait, device orientation.
- Edge compute
- TinyML inference on the low-power subsystem. Cloud is optional.
Held, not operated.
Grip, orientation, and contact are continuous inputs, not incidental. The sensor placement follows from how the device is held.
Device held naturally in one hand. Realistic hand scale and ergonomics.
Stands on its own.
A rear kickstand holds the device upright on a flat surface, screen and front camera facing the room.
Kickstand open. Concept render.
The form factor is not a constraint on the AI. It is the AI's first input.
Industrial design: US D831,024, US D899,422, WIPO DM/082 488, HK 1300965.1M001, CN 201330241241.1
Full electrical and mechanical specifications available in the data room.
The twin in action
One system. Three moments.
The same system, expressed against a different lived context.
01Workday Prediction
- Detects
- Elevated stress, commute movement, morning schedule.
- Predicts
- Work mode.
- Prepares
- Launches productivity apps, restores prior work layout, suppresses entertainment notifications.
02Fatigue Detection
- Detects
- Slowed interaction speed, physiological fatigue signal, reduced responsiveness.
- Predicts
- Cognitive load ceiling reached.
- Prepares
- Reduces notifications, dims the UI, delays non-urgent tasks.
03Creative Workflow
- Detects
- Music app active, stylus engaged, historical design workflow pattern.
- Predicts
- Creative session beginning.
- Prepares
- Restores the creative workspace, preloads editing tools, reopens reference materials.
Control and privacy
Predictive, without surrendering control.
Prediction this personal only works if the user stays in control.
When it acts, when it asks, when it stays silent
- 0–60%
Silent learning
Observe only. Store the pattern, interrupt nothing.
- 60–80%
Suggestion
Surface a non-intrusive suggestion the user can ignore.
- 80–95%
Soft action
Act, with an easy undo or an optional confirmation.
- 95%+
Automatic
Execute frictionlessly — but only for routine, low-risk actions.
The risk gate
High-risk actions always require approval, regardless of confidence.
Confidence is not the only test. High-risk actions — sending money, posting publicly, sharing sensitive data — never execute automatically, even at 99% confidence. They always ask.
Consistent + low-risk + frequently repeated → act automatically.
What stays, what leaves
Stays on device
- Raw PPG, GSR, and contour-sensor traces
- Raw camera and interface imagery
- Real-time sensing and state estimation
- Local PAS prediction and CascadeOS execution
Leaves, abstracted only
- Derived features: heart-rate trend class, HRV summary, stress-state embedding
- Structured interface descriptors, never screenshots
- Privacy-filtered event summaries
Sensitive data is processed on-device whenever possible. Only abstractions leave.
CascadeOS is the policy-enforcement layer that governs what leaves the device versus what stays local, on a process-first, transmit-later basis.
The opportunity
A platform at the convergence of AI, operating systems, and hardware.
Sentient sits at the convergence of predictive intelligence, the operating system, and consumer hardware. The edge AI software market below is the beachhead into a trillion-dollar opportunity.
Beachhead market
Edge AI software market
2025$40B
2030$110B
Source: ABI Research, 20251
Platform opportunity
$1T+
Total addressable market
Company-stated. Not a third-party estimate.
The wider platform opportunity spans adjacent markets. These are not presented as immediately addressable.
- Consumer hardware
- Mobile computing
- Enterprise platforms
- Gaming
- Healthcare
- Automotive
- Robotics
- Spatial computing
- Military technology
How it makes money
Three primary revenue categories at launch.
- 01
PAS DEVICES
$699 to $7,000 per device, consumer through enterprise tiers.
- 02
PAS SUBSCRIPTIONS
Per-user predictive intelligence, billed monthly.
- 03
CASCADEOS LICENSING
Per-device royalties and enterprise contracts.
Cloud intelligence APIs, enterprise and healthcare, developer platform, automotive and robotics, AI memory and digital twins, predictive commerce, and contextual advertising complete the ten-stream model. Detailed in the investor deck and data room.
Why it is defensible
- 01
Intellectual property
13 patents filed: 10 granted (1 utility, 9 design) and 3 pending (1 utility, 2 design), covering software and industrial design.
- 02
Vertically integrated architecture
PAS, CascadeOS, and SmartSelfie hardware documented as one vertically integrated stack.
- 03
Longitudinal personalization
Every device builds a behavioral digital twin of its user. Each twin compounds in per-user value the longer the device is in use.
The moat is not network effects on a social graph. It is longitudinal data depth on the individual.
ABI Research, Edge-Focused AI Chip Vendors Expand Their Software Stack (2025).
Traction and roadmap
What exists today, and what this round builds.
The IP, architecture, and raise plan are documented today. This round funds the path from granted patents to a device in market.
In hand today
IP portfolio
13 patents filed: 10 granted (1 utility, 9 design) and 3 pending (1 utility, 2 design), cited in the source materials.
System architecture
PAS, CascadeOS, and SmartSelfie hardware documented as one vertically integrated stack.
Founder dossier
Inventor track record, corporate background, and prior mobile-video IP work.
Investor materials
Business plan, financial model, and raise terms assembled and ready for the data room.
The plan
- Phase 1Year 1
Build
- PAS MVP
- AOSP implementation
- Initial SmartSelfie prototype
- First OEM discussions
- Phase 2Year 2
Validate
- Consumer beta
- Enterprise pilots
- Developer platform
- Phase 3Year 3
Commercialize
- Commercial launch
- OEM licensing
- Healthcare and enterprise expansion
The full five-year operating plan and revenue trajectory are in the investor materials.
Use of funds
- Engineering40%$20M
PAS, CascadeOS, SmartSelfie, Edge AI.
- AI Infrastructure15%$7.5M
Cloud, training, inference.
- Manufacturing & Hardware15%$7.5M
Prototypes, ODM relationships, sensor systems.
- Product & Design10%$5M
Product and design.
- Business Development10%$5M
OEM partnerships, enterprise sales.
- Legal & IP5%$2.5M
Patents, licensing, regulatory.
- Working Capital5%$2.5M
Day-to-day operations.
Figures are approximate. Full budget available in the data room.
Leadership
The people building it.
Franz Wakefield
Founder & CEO
Inventor and serial entrepreneur. Chemical Engineering, University of Florida. Prior roles at Dow, Kraft, and Johnson & Johnson.
Stephen Benedict
Chief Technology Officer
Electrical Engineering and MSE in Computer Science, University of Michigan. MBA, Harvard Business School. Prior technology leadership at Apple and Time Warner.
Benjamin Marouski
Chief Marketing Officer
Animation Director. Animation degree, Full Sail University. Expert in 3D rendering, polygonal and NURBS modeling.
The round.
The full deck, financial model, and data room are available to accredited investors on request.
Confidential. For accredited investor review only.
- Round
- $50M equity
- Runway
- 36 months
- Target
- $100M ARR