Dossi
Where insulin dosing meets predictive intelligence.
Dossi is an AI-powered insulin delivery system for Type 1 Diabetes. It considers multiple contextual factors — sleep, exercise, meal composition, site age, hormones — and learns your unique metabolic patterns over time. Not just what happened, but why.
Solo-designed and solo-built from the ground up. Currently functional and delivering insulin on my phone daily.
hellodossi.comThe Pitch
I have Type 1 Diabetes. I was diagnosed at age 14 and have been managing insulin delivery every day since. I saw a gap in the technology that no one was filling.
Current pumps only see glucose and carbs. I wanted something that saw the whole picture.
Before & After
A complete redesign from early prototype. Every screen was designed intentionally for warmth and ease of use. Data density increased while perceived complexity went down, creating a more efficient and approachable experience for users.
Design Process
Sign-Up Page
Five directions exploring tone, hierarchy, and how to balance warmth with medical credibility.
Onboarding Flow
Every screen was designed in Figma following a central design system.
App Icon
Design and development happened in parallel. The icon evolved alongside the product. Each build informed the next round of visual exploration.
16 icon directions tested gradient treatments, orb forms, and wordmark placement at every size. Each visual component of this app underwent a similar thoughtful process.
Soft purple gradients and organic shapes became the foundation of Dossi’s visual language, encouraging warmth and friendliness in contrast to typical cold medical tech aesthetic.
Not just what happened, but why.
How It Works
Continuous Monitoring
Reads glucose every 5 minutes via Dexcom Bluetooth. Syncs sleep, exercise, and heart rate from Apple HealthKit — no manual logging.
Contextual Learning
A Bayesian engine learns your individual patterns across sleep quality, infusion site age, dawn phenomenon, menstrual cycle, exercise, and caffeine.
Predictive Intelligence
Models glucose trajectories 4 hours ahead. Quantifies uncertainty. Adjusts insulin delivery in real time through closed-loop control.
AI Nutrition
Snap a photo of your meal and Dossi identifies the food, estimates macros, and predicts glucose impact — replacing manual carb counting with computer vision.
Key Screens
By the Numbers
Lines of Swift
Swift Files
Build Timeline
Developer
Bayesian Learning
Learns individual glucose patterns across 5+ contextual factors. Only applies effects after 15+ observations with 50%+ confidence. Baseline drift detection triggers recalibration.
BLE Pump Driver
Native Bluetooth LE driver for Omnipod DASH. Pod pairing, encrypted sessions, status monitoring, basal adjustments, and bolus delivery — all from scratch.
5-Layer Safety
Hard-coded limits, physiological bounds, hypo prediction, anomaly detection, and immutable audit logging. Biometric auth required for every dose. TOCTOU protection at delivery time.
AI Nutrition
Photo meal recognition via Gemini API. Snap a photo, identify foods, estimate macros, and predict glucose impact — no manual carb counting.
Closed-Loop Control
Model predictive control optimizes dosing decisions continuously. Reads glucose every 5 minutes and adjusts basal rates in real time.
Full-Stack
SwiftUI + SwiftData locally, Supabase for auth and cloud sync, WidgetKit for home screen, watchOS companion app. Swift 6 strict concurrency throughout.
Programs & Research
Factor Research
Catalogued 50 factors that affect blood sugar — from insulin timing to sleep quality to menstrual cycle phase. This research directly informed Dossi’s Bayesian learning engine.
| Factor | Category | Mechanism |
|---|---|---|
| Active energy burned | Activity | Load/intensity proxy |
| Aerobic exercise | Activity | Improves sensitivity |
| Alcohol intake | Nutrition | Hepatic suppression |
| Altitude | Environment | Stress/dehydration |
| Anaerobic / HIIT | Activity | Adrenaline response |
| Basal insulin delivered | Insulin delivery | Background glucose disposal |
| Basal rate changes | Insulin delivery | Time-varying background |
| Body temperature | Physiology | Fever / cycle proxy |
| Bolus insulin | Insulin delivery | Offsets meal glucose rise |
| Bolus stacking | Insulin (derived) | Overlapping doses |
| Caffeine | Nutrition | Catecholamine release |
| Carbohydrates | Nutrition | Primary glucose input |
| CGM glucose readings | Glucose | Observed interstitial glucose |
| CGM sensor lag | Glucose measurement | Interstitial vs blood delay |
| Circadian misalignment | Sleep | Hormonal rhythm shift |
| Cold exposure | Environment | Stress response |
| Complex carbs | Nutrition | Slower absorption |
| Compression lows | Glucose measurement | Mechanical pressure artifact |
| Correction bolus | Insulin delivery | Lowers hyperglycemia |
| Cortisol (chronic stress) | Hormones / cycle | Hepatic glucose output |
| Extended / dual bolus | Insulin delivery | Distributes bolus over time |
| Fat (total) | Nutrition | Delayed gastric emptying |
| Fiber | Nutrition | Slows absorption |
| Glucose trend / rate | Glucose (derived) | Slope / acceleration |
| Growth hormone overnight | Hormones / cycle | Nighttime resistance |
| Heart rate | Physiology | Stress/intensity proxy |
| Heart rate variability | Physiology | Recovery/stress proxy |
| Heat exposure | Environment | Increased absorption |
| Infection / illness | Illness | Inflammation-driven resistance |
| Infusion site aging | Pump / set | Reduced absorption |
| Insulin on Board | Insulin (derived) | Active insulin remaining |
| Late bolus timing | Insulin (derived) | Carbs before insulin |
| Meal timing | Nutrition | Insulin–carb alignment |
| Menstrual follicular | Hormones / cycle | Increased sensitivity |
| Menstrual luteal | Hormones / cycle | Increased resistance |
| Missed bolus | Behavior | Omitted insulin |
| Occlusion / kink / leak | Pump / set | Failed delivery |
| Post-exercise window | Activity | Enhanced insulin action |
| Protein | Nutrition | Gluconeogenesis |
| Recovery after illness | Illness | Sensitivity rebound |
| Simple carbs | Nutrition | Fast absorption |
| Sleep duration | Sleep | Sleep debt increases resistance |
| Sleep quality | Sleep | Poor recovery |
| Steps / movement | Activity | Mild uptake |
| Suspend / pause delivery | Insulin delivery | Stops insulin flow |
| Systemic steroids | Medications | Strong insulin resistance |
| Temp basal decrease | Insulin delivery | Short-term reduction |
| Temp basal increase | Insulin delivery | Short-term increase |
| Water / hydration | Nutrition | Indirect physiology |
| Workout session | Activity | Increases uptake |
Programs
Built through Georgia Tech’s InVenture Prize and Startup Exchange. Mentored by Rosa Arriaga, Associate Professor in the School of Interactive Computing.
User Research
Distributed flyers through Georgia Tech Disability Services to survey T1D students. The most consistent feedback: people wanted to understand patterns, not just numbers.
