Dossi

AI-guided insulin dosing for Type 1 Diabetes.

Dossi Dashboard

About Me

Hi, I'm Casey

Georgia Tech, Industrial Design
I want to be an entrepreneur
Currently in Idea to Prototype (I2P)
I have Type 1 Diabetes
Former Breakthrough T1D Youth Ambassador
Omnipod insulin pump

The Problem

My "Smart" Pump Isn't Very Smart

I wear an insulin pump and a continuous glucose monitor.

But it really only knows two things:

1
My current glucose
2
Carbs I'm about to eat

Background

Two Types of Insulin Delivery

Managing Type 1 diabetes requires balancing two distinct insulin strategies:

Basal Insulin

Continuous background delivery

  • Runs 24/7 at varying rates
  • Keeps glucose stable between meals
  • Adjusts for sleep, activity, time of day

Bolus Insulin

On-demand meal-time doses

  • Covers carbs you're about to eat
  • Corrects high glucose readings
  • Varies based on food type and timing

Current pumps handle both — but only with basic rules. Dossi makes both smarter.

The Gap

What My Pump Doesn't Know

  • I ran 2 miles this morning — I'm more sensitive for the morning
  • I slept 4 hours last night — My glucose will run higher
  • I'm eating pizza, not salad — Fat causes delayed spike 4 hours later
  • I'm getting sick — My needs are about to double

I adjust for all of this manually. Every day.

The Solution

An insulin advisor that sees the full picture

Dossi considers 40+ factors and learns your patterns over time.

Context Awareness

The 40+ Factors

GlucoseValue, trend, velocity, momentum, quality, staleness
InsulinIOB, decay curve, recent boluses, basal rate, type
NutritionCarbs, protein, fat, fiber, absorption, meal type
ActivityWorkouts, time since, intensity, steps, heart rate
Sleep & BodyHours, quality, debt, cycle phase, illness, stress
EnvironmentTime, day, temperature, site age, location, absorption

How It Works

Getting Started

Personalized From Day One

During onboarding, Dossi pulls your historical health data from phone records.

PatternWhat It Learns
Dawn PhenomenonYour morning glucose rise timing & magnitude
Exercise SensitivityHow workouts affect your insulin needs
Basal PatternsYour hourly insulin requirements
Meal ResponseYour food absorption rates by meal type

With 14+ days of data, machine learning detects your patterns.

How It Works

Step 1: Photo Meal Recognition

1.Snap a photo of any meal
2.AI identifies individual foods
3.Estimates carbs, protein, fat, fiber
4.Models interaction of different macros
5.Suggests dosage
Nutrition Tracker

How It Works

Step 2: Context Assembly

Every time you log a meal, Dossi pulls data from multiple sources in parallel:

HealthKit
CGM
Sleep
Activity
History
Time

The result: a complete snapshot of your current state in under 1 second.

Learning loop

How It Works

Step 3: The Learning Loop

After every meal, Dossi observes what actually happened:

1.
Predict — Given this context, glucose should land at X
2.
Observe — Actual glucose was Y (higher or lower?)
3.
Attribute — Which factors were active?
4.
Update — Adjust personal sensitivity model for that factor

Beyond Meals

Basal Rate Intelligence

Dossi learns your background insulin needs throughout the day and automatically adjusts delivery.

12am – 4am
Deep Sleep
Lower rates during rest
4am – 8am
Dawn Phenomenon
Hormones spike glucose
8am – 12pm
Morning Active
Moderate baseline
12pm – 6pm
Afternoon
Activity-adjusted
6pm – 12am
Evening Wind-down
Gradual reduction

Dossi learns YOUR unique patterns and suggests basal adjustments automatically.

Safety First

5 Layers of Validation

Insulin errors can be dangerous. Every recommendation passes through graduated safety checks.

1
Hard Limits — Max 25U bolus, min 70 mg/dL glucose, max fall rate
2
Physiological — Prevents insulin stacking and unrealistic doses
3
Predictive — Calculates hypo risk at 15/30/60/120 min horizons
4
Anomaly Detection — Validates sensor data quality and flags issues
5
Learning Validation — Checks recent outcomes for algorithm drift

The Impact

Why This Matters

1.7M
Americans live with Type 1 diabetes.
21%
Only 21% of adults with T1D meet glucose targets.
180
Extra decisions made per day by people with T1D

Business Model

Path to Market

The global diabetes care market is valued at $72 billion.
Insulin delivery devices are growing 8% annually.

Regulatory FDA 510(k) clearance as Class II medical device
Distribution Prescription app sold through healthcare providers
Coverage Insurance reimbursement under diabetes management codes
Pricing $30-50/month subscription or annual plans
FDA logo Tidepool logo Bigfoot Biomedical logo

Tidepool and Bigfoot have proven this regulatory and reimbursement model.

Motivation

Why I'm Doing This

  • I live this problem — I know exactly what's missing
  • The tech is ready — HealthKit, Apple Watch, FDA pathway proven
  • Nobody else is doing this — Still stuck on glucose + carbs

Roadmap

What's Next

Now
Finish build
Self-testing
Validate algorithm
Summer
CREATE-X
Pilot with T1D users
Conduct research
Future
FDA clearance
Direct pump control
Push mainstream

Mentorship

Where I'd Love Guidance

1.
Navigating medical design
2.
Feedback on my work and suggest improvements
3.
Out of the idea realm
4.
Possible connections to others who could help me

Thank you!