Physical AI-powered digital twin for anxiety reduction
Invalid video URL
Overview
PART OF MIT HARD MODE - HARDWARE + AI HACKATHON Winner of the "Thrive" category JIRA started with a simple observation: travel anxiety is invisible, isolating, and completely underserved by anything on the market. Millions of people fly every year gripping their armrests with nothing but a pill or a pair of headphones to get them through it, and neither actually responds to them. JIRA does. It's a physical AI cat companion built into a travel pillow, with a personality modeled on real feline behavior and driven by live sensor data. Pet its head and it warms to you. Touch its tail and it gets annoyed. When it's happy, it purrs, using haptic vibrations tuned to the 20–140 Hz range, which has been shown to lower cortisol and reduce blood pressure. Its tail is a full emotional output channel, shifting between a slow flick of contentment, a rapid thump of irritation, and a held stillness that signals attention. Under the hood, the personality engine runs on Claude by Anthropic, which means JIRA reasons about context rather than just reacting to input. A white-knuckle grip during turbulence lands differently than a gentle tap in calm air, and the response shifts accordingly. You can talk to it, vent at it, or just hold it in silence and let it figure out what you need.
Roles
Client / Creative Collective
Massachusetts Institute of Technology
Collaborators
Featured Photos
What it does
JIRA is a physical AI cat companion built into a travel pillow. Its personality is modeled on real feline behavior and dynamically generated from live sensor data. Pet its head and it responds with warmth. Touch its tail and it gets annoyed. When it is genuinely happy, it purrs. The vibration motors deliver haptic feedback tuned to the 20 to 140 Hz range shown in research to lower cortisol and reduce blood pressure. The tail is a full emotional output channel. Multiple mechanical modes express distinct states: a slow flick of contentment, a rapid thump of irritation, a held stillness that signals attention. Users communicate through touch and motion. JIRA communicates back through vocalization, vibration, and movement. The personality engine runs on Claude, Anthropic's AI. Claude reasons about context rather than just reacting to input. A tight grip during turbulence means something different than a light tap in calm air. The response shifts accordingly. JIRA can also hold a real conversation. You can talk to it, vent at it, or hold it in silence and let it decide what you need.
How we built it
The hardware consists of vibration motors for purring simulation, a subwoofer for low-frequency resonance, a motorized tail with multiple mechanical movement modes, an IMU for reading motion context like turbulence and restlessness, and pressure sensors that register contact and grip intensity. All of this is packed into a travel pillow form factor designed to be held. The sensor data feeds a personality engine powered by Claude via the Anthropic API. Real-time readings from the IMU and pressure sensors drive the model's context window, so the AI's responses are grounded in what the user is physically experiencing at that moment. The output maps back to hardware: vocalization patterns, vibration intensity, and tail movement mode.
Challenges we ran into
Mapping physical sensor data to emotionally meaningful AI responses in real time required careful prompt engineering. The latency between a sensor event and a physical hardware response needed to stay low enough to feel natural, not like a system thinking. Mechanical tail design was also a significant constraint: expressing a meaningful range of emotional states through a single actuated appendage, reliably, in a compact form factor, took many iterations.
Accomplishments that we're proud of
The personality feels real. The small moments of feline resistance, the earned affection, the unpredictability of a cat that has opinions about where you touch it, these are what make JIRA feel like a companion rather than a device. We are proud that the "annoyed tail" was not cut from the product. It is one of the most important features we built. We are also proud of the therapeutic framing. The subwoofer is not an audio gimmick. The purr frequency range has clinical backing. We built a delivery mechanism for something that genuinely helps people, and disguised it as something you just want to hold.
What we learned and what's next
Designing for emotional response is harder than designing for functional response. A feature that works correctly can still feel wrong. We learned to prototype the feeling first and the mechanism second. We also learned that Claude's ability to reason about context, rather than simply pattern-match to inputs, was essential to making the personality feel coherent across a long flight rather than just in isolated interactions. We want to expand the personality model to learn individual users over time. Your JIRA should know that you grip tighter on takeoff than landing, and adjust accordingly. We are also exploring biometric integration, heart rate and skin conductance, to give the AI an even richer picture of the user's state. Longer term, JIRA should not be limited to travel. Anywhere anxiety lives, JIRA should be able to go.