Under the Hood
Nearly a decade of biomechanical research, behavioral modeling, and real-time AI engineering — markerless pose, a stateful kinematic engine, self-supervised vision fusion, and a pet that speaks in its own personality — compressed into an experience that fits in your pocket and responds in about a second.
Every second your phone's camera is pointed at your pet, Como is running a four-stage intelligence pipeline — from raw, uncalibrated video to an adopted persona speaking in real time.
Como starts where laboratory biomechanics ends: an ordinary phone, hand-held, indoors, uncontrolled lighting, a pet that won't hold still. The Como Markerless Pose Engine (CMPE) was built for exactly this — recovering a full, confidence-annotated skeleton from in-the-wild consumer video, where laboratory-grade motion capture is impossible.
CMPE detects 20 anatomical keypoints across your pet's body — eyes, ears, throat, withers, forelimb joints and paws, hindlimb joints and paws, and tail set — each with a live confidence value, every frame. The result is a real-time skeletal map that survives partial occlusion, awkward camera angles, and moving subjects, and always knows exactly what it can trust.
The live skeleton flows into a stateful Kinematic Engine computing six gated locomotion traits over device-invariant 3-second windows — gait symmetry, weight distribution, posture, and more — with a three-tier confidence model and a flag-gating cascade that refuses to guess when the data doesn't support it. In parallel, a self-supervised two-encoder network (temporal pose transformer + vision transformer, served entirely via ONNX) classifies behavior — with no third-party vision model in the loop.
The analysis flows into a synthesized personality that speaks as your pet — fused from six evidence streams and expressed through a persistent, low-latency voice session. In about a second, Como speaks. Not a pre-recorded phrase. A contextually aware statement about the specific animal in front of you — in their personality, in their voice.
Como tracks 20+ anatomical keypoints per video frame — gait, posture, movement quality, energy — and compares them against a behavioral baseline built for your pet alone, not a generic average.
The methodology behind Como has been reviewed and audited by leading world professors, and every layer of the pipeline is built exclusively on peer-reviewed research — published biomechanics literature, the Glasgow CMPS-SF pain scale, SNOMED-CT clinical terminology.
What we can say: Como never gets tired, never has competing information, and analyzes thousands of data points per second across dozens of body regions — then speaks, in about a second.
Como's methodology has been reviewed and audited by leading world professors and is built exclusively on peer-reviewed research. The core technology is protected by 11 patent families, including a patent application already approved in Canada. Como is a health-and-behavior insight tool, not a veterinary device: it prepares findings to share with a licensed veterinarian, not replace one.
Pain scoring is validated against the Glasgow CMPS-SF composite measure and coded with SNOMED-CT clinical terminology. Every pain and risk output is continuously probed for demographic bias — the same behavioral input must produce byte-identical output across every group, with zero tolerance for deviation. 51 / 51 tests passing.
Como's voice layer runs on a persistent, low-latency bidirectional audio protocol — not a request-response cycle, not a chatbot. A true continuous conversation grounded in live biomechanics, fused vision, and everything the pet has ever shown you.
And Como doesn't talk about your pet. It talks as your pet — through a synthesized personality built from who they actually are. When your dog's anxiety markers spike, Como doesn't just report it. It responds in a voice that sounds like it genuinely knows them. Because it does.
Perception, memory, and reasoning never change. What changes is who's speaking. Como fuses six evidence streams — a 13-question owner profile, breed-science temperament priors, the pet's own kinematic history, AI-derived behavioral insights, your conversation history, and long-term memory — into a structured persona document. The conversational agent doesn't narrate your pet. It adopts the pet as itself — and you choose the personality it speaks through.
Clinical veterinary voice — gait, posture, pain assessment, and structured reports you can hand to a real vet.
Simple, educational, wholesome. Explains your pet to the smallest members of the family.
Narrative, poetic, imaginative. Every walk becomes a chapter.
Comedy, jokes, sarcastic wit. Your pet is the comedian.
Classic Disney quotes and phrases, woven into every observation.
Exactly like Dug from Pixar's UP. Squirrel!
Wise, poetic, quotes literature. Your pet as a well-read philosopher.
Paranoid, suspicious, connecting dots. Nothing is a coincidence.
Chaotic, uncensored, wild. For owners with a sense of humor.
From a clinical veterinary assistant to full comedy — it's the same engine. The same 20-keypoint skeleton, the same kinematic reasoning, the same memory. Only the persona changes.
Como runs a dual memory system. A live session memory captures everything happening right now — conversation, emotional state, behavioral events, camera observations. A long-term memory accumulates across every session, every week, every year — including a biomechanical episodic store, so your pet's movement history is remembered as vividly as its words.
Real-time context: the full conversation, emotional arc, behavioral events, and camera observations — recalled in under a second.
Months of health events, behavioral patterns, emotional trends, and training milestones — every session makes Como deeper, not just faster.
Your pet's movement history, stored as a 12-dimensional kinematic embedding — enabling trend reasoning: "is this limp worse than last week?"
Consistency is the single most important variable in pet health outcomes. It's also the one most pet owners struggle with. We studied the behavioral mechanics behind the most effective habit-formation systems ever built — and applied them to the one motivation that is genuinely immune to rationalization: the wellbeing of an animal you love.
XP. Levels. Streaks. Daily challenges. The gamification layer isn't decorative — it's the mechanism that converts good intentions into measurable behavioral change. In our early users, daily care consistency increased by over 3× within the first two weeks.