MIOM
A gut-health companion for people who move abroad. A wearable sensor, a voice-first AI agent, and a communal dock help them see how their gut is adjusting and compare notes with others making the same move.
Core Concept
MIOM is the only relocation companion that turns personal environmental data into shared community knowledge, stored locally and open about its sources. It links the symptoms people notice after moving abroad to the biology behind them.
My Role
- Research: user and expert interviews, synthesis into the seven-factor model
- UX and system journeys: the three-stage journey, modes, and device status behaviours across all three objects
- AI agent and voice design: personality, knowledge registers, dialogue, and the voice interaction states
- Working app prototype: mobile app, agent backend, and generative biome portrait
- My teammates led the physical form of the device and dock
Your gut microbiome relocates with you
Every year, millions of people move abroad for study or work. Almost no one tells them that their gut microbiome moves with them, and that it will spend months adjusting to a new food system, water profile, climate, and the compound stress of starting over.
The disruption shows up in the body: stomach issues in the first weeks, low energy that does not resolve, recurring colds, skin reactions, poor sleep. Because each symptom appears separately, people blame each one on something different: the flight, the stress, the new dairy. Few of them connect it all to a gut adjusting to a very different environment. It is a health literacy gap rather than a medical condition, and no product currently helps people understand it.
“With many changes to your environment you just brush to the side… but it's a lot when you think about it.” Interview participant, first month after relocating
Problem Statement
How might we help people who relocate make sense of what their body is going through, by connecting and translating what they notice, without diagnosing or prescribing?
Established science, unexplained experience
Vangay et al. (Cell, 2018) tracked refugees relocating from Southeast Asia to the United States and found they began losing native gut microbes almost immediately after arrival, with microbial diversity declining for at least ten years. Relocation changes seven environmental factors at once, and research links each of them to gut disruption on its own.
We interviewed international students and expats who had relocated from Taiwan, Turkey, India, the United States, Panama, Australia, and Norway, alongside experts in gut health, internal medicine, public health, and geomorphology. The pattern was consistent: symptoms were noticed in isolation and attributed to single, surface-level causes. Nobody had a framework connecting them.
- Food systems and antibiotic residues
- Psychological stress
- Social environment change
- Air quality
- Water quality
- Climate, temperature, and nature exposure
- Sleep and circadian disruption
Key insights
- People notice their symptoms but don't connect them. So the design job is translating what they already feel; detecting it is not the gap.
- Food is the easiest way in. People already notice, photograph, and talk about what they eat, even though food is only one of the seven factors.
- Noticing is useful by itself. Photographing a meal prompts reflection before any analysis happens.
- Water quality is underestimated. For many people it disrupts the body sooner than food does.
- Individual experience needs a collective mirror. Learning that others feel the same thing reassured people more than anything else.
- People trust what they can follow. The system has to show where each piece of knowledge comes from.
Three objects, one loop of mutual learning
Individual observation → personal sense-making → community contribution → richer collective knowledge → better sense-making for everyone.
The portable device
A small companion carried on a bag or keychain. It passively senses temperature and humidity through the day and actively captures food images with a press of its button. Its round screen shows a simplified biome portrait of the day's traces.
The digital companion
A mobile app housing the AI agent, also named Miom. Conversations happen here, and the adaptation journey is visualised over time as a living data portrait. The agent draws on three knowledge sources: the user's personal data, community knowledge pooled from others in the same transition, and the scientific evidence base.
The communal dock
A shared station in university common spaces. Docking moves data in both directions: personal traces join the community pool, and the accumulated knowledge of everyone who has docked flows back to the device. The dock displays a living visualisation of the collective experience and speaks ambient insights aloud to whoever is nearby.
Mapping every moment across three objects and three modalities
I mapped the experience as three stages, tagging every step with the object it happens on (device, dock, or app) and the interface mode it uses: tangible (TUI), graphical (GUI), or voice (VUI). The map made it possible to give each object a strict role: the device notices, the app explains, the dock connects.
From a dashboard to a data portrait
The project started under the working name Biome. Early on we decided the visualisation should work like an artwork, so people sense their data instead of reading numbers off a chart. I prototyped the portrait as a generative petri dish whose colonies grow with days logged, cluster by food diversity, and fray at the edges with disruption, tuning it live through a parameter panel before committing to a visual language.
Three interactions, one companion
The app is built around three moments, each with its own trigger and its own voice. Every clip below is a screen recording of the working prototype.
Onboarding: an agent that doesn't know you yet
Miom introduces itself and collects the baseline it needs before it can say anything specific: where you came from, where you are now, what you used to eat, and how you feel. It infers what it can (“I noticed that you are currently in Lugano”) and asks one question at a time. The conversation is voice-first, and text input is always there.
Reflection: triggered by touch, never by a menu
Reflection opens only when the user taps the device to the phone (“Miom tapped”). Miom surfaces one pattern it chose from your own data and shows why: a becausechip cites the evidence, such as “3 of your last 5 pasta meals were followed by a bloating note.”
Ask Miom: on your terms
The user can open Ask Miom whenever they have a question. Here Miom is direct and brief. It answers from personal data, community knowledge, and research, and says which one it is using.
The biome portrait
The home screen and the device share one generative portrait, rebuilt from the user's data. It starts sparse on day zero and fills as meals, readings, and community knowledge accumulate.
A companion, not an assistant
An assistant answers questions and executes tasks. A companion notices things, asks questions back, and is curious. Miom should feel like the friend who knows about this stuff, the one you call after a confusing week in a new country: warm, specific, honest about uncertainty, and never authoritative. It gives context and asks questions, and the user decides what to do.
What Miom never does
- Diagnose symptoms or suggest what a user might have.
- Recommend medications, supplements, or clinical treatments.
- Substitute for medical care or professional advice.
- Offer unsolicited health guidance.
- Reference any individual's data in a public or communal context.
- Speak about theoretical knowledge as if it were established fact.
Voice states
The voice orb gives the agent a body. Each state is distinct at a glance, so the user always knows whether Miom is waiting, hearing them, working, or speaking.
Three knowledge registers
How sure the agent sounds depends on where its knowledge comes from, so calibrating confidence became part of designing its personality.
Personal data: assertive, specific
Community: assertive, collective
Research: speculative, careful
Miom:I've been looking at what you've been eating this week. A lot of bread, pasta, some dairy, quite a change from what you described at home. You mentioned on Thursday that you felt heavy after lunch. Does that happen often?
User: Pretty much every day.
Miom:That makes sense. Your gut has been used to a very different daily input. What you're eating here is nutritionally fine, but the bacterial landscape is quite different. It takes a few weeks for things to start settling. You're not doing anything wrong.
Dock mode
On the dock, Miom changes register. It speaks in the third person about collective experience to whoever is in the room, so it sounds like part of the space.
“In the first few weeks, eating alone is one of the most common things people mention here. That tends to shift as people find their rhythm.”
Legible without looking closely
The portable device communicates through three output channels: its round screen, an LED, and vibration. I designed each state so it can be told apart by feel and peripheral vision alone.
| State | Screen | Light | Vibration | Meaning |
|---|---|---|---|---|
| Idle, passive sensing | Simplified biome portrait | Soft ambient pulse | None | On and sensing continuously |
| Active capture | Live camera viewfinder | Steady | None | Food photo capture started |
| Capture confirmed | Brief flash animation | Flash | Single pulse | Photo taken and saved |
| Bluetooth syncing | Subtle animation | Faint blink | None | Syncing to phone in the background |
| NFC reflection ready | Reflection prompt | Steady | Single pulse | Touching the phone; reflection can begin |
| Docking in progress | Community transfer animation | Sequence | Double pulse | Data moving to and from the dock |
| Docking complete | Community biome portrait | Steady warm glow | Long single pulse | Community data received |
| Low battery | Battery indicator | Slow red blink | Three pulses | Charge required |
A working system built in code
I built the app and the agent backend with Claude Code, starting API-first so the logic and behaviour could be tested before any visual polish, then refining the UI in Figma and syncing it back into the code through the Figma MCP. A Node backend routes each surface to its mode (onboarding, reflection, on demand, or dock), assembles a context bundle from the user's data, and composes the system prompt from a shared persona plus a mode layer. Gemini handles text and real-time voice. The device and dock run on Arduino UNO Q boards that reach the backend over a Tailscale VPN.
In the concept, personal data stays on the phone and device and community data stays inside the institution's dock. For the prototype, Supabase stands in for that local storage so the demo could run across devices, with its tables organised by the same three knowledge registers the agent speaks in.
Transparency as an interaction pattern
An AI health product has to balance assistance against autonomy, transparency against complexity, and individual benefit against collective responsibility. Miom handles these with five interaction patterns.
- Confidence calibration by knowledge source. The agent signals through language where each insight comes from.
- Evidence on screen. Reflection insights carry a “because” chip citing the data behind them.
- Active consent for community sharing. Nothing joins the community pool without an explicit choice and a physical act of docking.
- Local data storage. Personal data lives on the phone and device; community data lives in the institution's dock.
- Assistance without overreach. The system explains patterns and shares experiences instead of prescribing, and directs medical concerns to professionals.
Testing the system in the body and in space
Before any final form, we bodystormed the system with rough clay and cardboard stand-ins: carrying the device through a day, capturing a meal, and trying two dock orientations in a shared space.
My teammates then led the physical form, moving from clay and styrofoam studies to a parametric model in Fusion 360 and real-scale 3D prints that tested grip, screen and button placement, and how the device seats into the dock.
