Nachiket Sahare - Creative Designer

Home

Resume

Managing Diabetes, Not Just Tracking Numbers

An all-in-one diabetes app that reduces daily management burden through AI food recognition (92% accuracy), ingredient-based recipes, and seamless doctor collaboration—designed for real people, not just data points.

Diabetes apps fail 500M+ users by removing safety features, generating wrong calculations (28g → 50g+ errors), and excluding 65+ users through poor accessibility. 100% of interviewed users perform unsafe mental math for insulin doses with zero technological support.

As UX Lead, I defined and led the 0→1 redesign of safety-first diabetes management. I conducted comprehensive user research, built medical device-grade requirements (29 safety standards), and drove iterative prototyping from 62% to 85% coverage—resolving all critical usability issues and establishing the accessible UX foundation for diabetes management on mobile.

Product

Healthcare

Skills

Product Design

User Research & Testing

Interactive Prototyping

My Role

UX Lead

Timeline

Sept ‘25 - Dec ‘25

Tools

Figma, Dovetail, Figma Make, Claude

When Apps Fail, Lives Are at Risk

Diabetes is a life-critical condition affecting over 500 million people worldwide. Every day, people with diabetes make dozens of health decisions: How much insulin to take? What to eat? When to check glucose? These aren't casual choices—they're life-or-death calculations.

When I began researching the diabetes app market, I discovered a troubling reality: the apps designed to help people manage this condition were systematically failing them.

They just tell that don't eat this, but what to eat exactly that nobody tells clearly.

— User Research Participant, Type 2 Diabetes

No Dietary Guidance

Apps tell users what NOT to eat, but provide no actionable guidance on what TO eat

Missing Critical Features

70% of glucose excursions missed due to reactive (not proactive) monitoring approaches

Estimation Errors

±40% carbohydrate estimation errors lead to dangerous dosing miscalculations

The Design Challenge

How might we create a diabetes management interface that provides reliable data sync, intelligent dietary guidance, and actionable trend analysis—while remaining accessible to users of all ages?

Understanding Real User Contexts

Comprehensive mixed-methods research to understand user needs, pain points, and safety requirements

In-Depth Interviews

30-40 min each

200+

Insights

Thematic analysis

Apps Analyzed

Competitive analysis

FMEA

Risk Analysis

FDA standards

Journey

Mapping

Daily touchpoints

User Interviews

"When we were using a finger stick, we were actually getting different readings from the CGM versus what the finger stick was showing. I'm staring at two different numbers—142 on my device, 158 in the app—and I have to decide right now how much insulin to inject. Which one do I trust with my life?"

Through in-depth user interviews with 3 participants living with diabetes (ages 24-54, Type 1 and Type 2), I gained comprehensive insight into participants' daily struggles with glucose monitoring accuracy and data reliability. These insights revealed that trust—not features—was the fundamental barrier preventing users from relying on digital solutions for life-critical treatment decisions.

Key Findings

"When we were using a finger stick, we were actually note down the readings in an Excel on the computer manually. Which was like a very hectic process."

— Sahil, Type 1 (24 years old, 16 years with diabetes)

0:00/1:34

Insight: Apps must support multiple device types with robust error handling

"Someone is an old person with no knowledge of how graphs work... It is kind of difficult to navigate at start."

— Sahil (reflecting on older users)

0:00/1:34

Insight: WCAG 2.1 AA compliance is baseline for safety-critical features

"Maybe once in 15 days, or sometimes once in a month. When I feel that there is some discomfort, then I check."

— Arindam, Type 2

Insight: Shift from reactive logging to proactive monitoring with alerts

"They just tell that don't eat this, but what to eat exactly that nobody tells clearly."

— Nandini, Type 2 (54 years old, 19 years with diabetes)

Insight: Provide 'eat this instead' recommendations, not just restrictions

Meet Our Primary User

Understanding who we're solving for

Priya Sharma

"The Overwhelmed Seeker"

"They just tell that don't eat this, but what to eat exactly that nobody tells clearly. I need to know what's safe for me to eat, not just what to avoid. When I stand in my kitchen at lunchtime, I'm paralyzed by choices—the app gives me a list of 50 'forbidden' foods but zero suggestions for what I CAN eat."

Demographics

Age

54 years

Occupation

School Administrator

Location

Urban India

Tech Proficiency

Moderate

Diabetes Profile

Type

Type 2 Diabetes

Duration

19 years

Management

Medication + visits

Testing Frequency

Bi-weekly

Demographics

Age

54 years

Occupation

School Administrator

Location

Urban India

Tech Proficiency

Moderate

Diabetes Profile

Type

Type 2 Diabetes

Duration

19 years

Management

Medication + visits

Testing Frequency

Bi-weekly

Demographics

Age

54 years

Occupation

School Administrator

Location

Urban India

Tech Proficiency

Moderate

Diabetes Profile

Type

Type 2 Diabetes

Duration

19 years

Management

Medication + visits

Testing Frequency

Bi-weekly

Pain Points

Negative Guidance — Told what NOT to eat, no advice on what TO eat

Manual Entry Burden — Food logging too time-consuming

Small Fonts & Jargon — Apps difficult to navigate

Device Distrust — Different readings erode confidence

Goals

Find energy-sustaining foods for workday

Get positive dietary guidance

Avoid future complications

Top 3 Critical Needs

AI Food Recognition

Snap photos to log meals automatically

Positive Dietary Guidance

"Eat this instead" suggestions

Accessible Interface

Large fonts, intuitive navigation

Initial Prototype: Building the Foundation

Based on 34 FDA medical device software requirements identified through research and FMEA analysis, I created a high-fidelity interactive prototype in Figma using AI-assisted design tools (Figma Make). The initial prototype (v1.0) achieved 62% requirements coverage with 5 core features addressing the most critical user needs identified during research.

62% Requirements Coverage (v1.0)

15 Interactive Screens

4 Core Features Built

62% Requirements Coverage (v1.0)

15 Interactive Screens

4 Core Features Built

Core Featured Designed

Glucose Dashboard

Daily glucose visualization with 24-hour trend graph showing patterns and fluctuations at a glance.

24-hour color-coded trend graph (Green/Orange/Red zones)

Current glucose reading with status indicator

Date navigation to view historical days

Visual pattern recognition for daily trends

Meal Analysis & Tracking

Comprehensive meal logging with nutritional breakdown and historical tracking to understand personal glucose patterns.

Log meals with automatic timestamp capture

Meal composition analysis (carbs, protein, fat breakdown)

Photo attachment option for visual meal records

Historical meal tracking with glucose correlation insights

Meal Prediction & Suggestions

AI-powered food database with glucose impact predictions addressing the core user pain point: "What should I eat?”

Searchable food database (100,000+ items with nutrition info)

Predicted glucose response curve based on carbohydrate content

Eat this instead" alternative recommendations for better choices

Portion size calculator with visual reference guides

Medication Tracking & Reminders

Comprehensive medication management system with dose logging, smart reminders, and adherence tracking.

Medication list with name, dosage, and schedule

Smart dose reminders at scheduled times

Quick log actions (Taken/Skip/Snooze)

7-day adherence calendar view

Glucose Dashboard

Daily glucose visualization with 24-hour trend graph showing patterns and fluctuations at a glance.

24-hour color-coded trend graph (Green/Orange/Red zones)

Current glucose reading with status indicator

Date navigation to view historical days

Visual pattern recognition for daily trends

Meal Prediction & Suggestions

AI-powered food database with glucose impact predictions addressing the core user pain point: "What should I eat?”

Searchable food database (100,000+ items with nutrition info)

Predicted glucose response curve based on carbohydrate content

Eat this instead" alternative recommendations for better choices

Portion size calculator with visual reference guides

Meal Analysis & Tracking

Comprehensive meal logging with nutritional breakdown and historical tracking to understand personal glucose patterns.

Log meals with automatic timestamp capture

Meal composition analysis (carbs, protein, fat breakdown)

Photo attachment option for visual meal records

Historical meal tracking with glucose correlation insights

Medication Tracking & Reminders

Comprehensive medication management system with dose logging, smart reminders, and adherence tracking.

Medication list with name, dosage, and schedule

Smart dose reminders at scheduled times

Quick log actions (Taken/Skip/Snooze)

7-day adherence calendar view

Glucose Dashboard

Daily glucose visualization with 24-hour trend graph showing patterns and fluctuations at a glance.

24-hour color-coded trend graph (Green/Orange/Red zones)

Current glucose reading with status indicator

Date navigation to view historical days

Visual pattern recognition for daily trends

Meal Prediction & Suggestions

AI-powered food database with glucose impact predictions addressing the core user pain point: "What should I eat?”

Searchable food database (100,000+ items with nutrition info)

Predicted glucose response curve based on carbohydrate content

Eat this instead" alternative recommendations for better choices

Portion size calculator with visual reference guides

Meal Analysis & Tracking

Comprehensive meal logging with nutritional breakdown and historical tracking to understand personal glucose patterns.

Log meals with automatic timestamp capture

Meal composition analysis (carbs, protein, fat breakdown)

Photo attachment option for visual meal records

Historical meal tracking with glucose correlation insights

Medication Tracking & Reminders

Comprehensive medication management system with dose logging, smart reminders, and adherence tracking.

Medication list with name, dosage, and schedule

Smart dose reminders at scheduled times

Quick log actions (Taken/Skip/Snooze)

7-day adherence calendar view

Glucose Dashboard

Daily glucose visualization with 24-hour trend graph showing patterns and fluctuations at a glance.

24-hour color-coded trend graph (Green/Orange/Red zones)

Current glucose reading with status indicator

Date navigation to view historical days

Visual pattern recognition for daily trends

Meal Prediction & Suggestions

AI-powered food database with glucose impact predictions addressing the core user pain point: "What should I eat?”

Searchable food database (100,000+ items with nutrition info)

Predicted glucose response curve based on carbohydrate content

Eat this instead" alternative recommendations for better choices

Portion size calculator with visual reference guides

Meal Analysis & Tracking

Comprehensive meal logging with nutritional breakdown and historical tracking to understand personal glucose patterns.

Log meals with automatic timestamp capture

Meal composition analysis (carbs, protein, fat breakdown)

Photo attachment option for visual meal records

Historical meal tracking with glucose correlation insights

Medication Tracking & Reminders

Comprehensive medication management system with dose logging, smart reminders, and adherence tracking.

Medication list with name, dosage, and schedule

Smart dose reminders at scheduled times

Quick log actions (Taken/Skip/Snooze)

7-day adherence calendar view

Initial Prototype Screens

Concept 1: Dashboard with daily glucose overview

Concept 2: Meal Prediction & Suggestions

Concept 3: Meal Analysis & Tracking

Concept 4: Medication Tracking & Reminders

Initial Prototype Screens

62%

FDA compliance

Issues Found (AI Test)

Features Functional

91%

AI Assessment (Grade A)

Usability Testing & Iteration

November 18-20, 2025 • Discovering critical failures through real user testing

Testing Approach

I conducted moderated usability testing over three days (November 18-20, 2025) with 3 participants representing our target demographic. Each session lasted 45-60 minutes and included task-based testing with prototype v1.0, think-aloud protocol, and post-task interviews about pain points.

4 Task Flow

Remote testing using GMeet

AI-assisted analysis using Dovetail

4 Task Flow

AI-assisted analysis using Dovetail

Remote testing using GMeet

4 Task Flow

Remote testing using GMeet

AI-assisted analysis using Dovetail

4 Task Flow

Remote testing using GMeet

AI-assisted analysis using Dovetail

Overall Performance

100%

Task Success Rate (12/12)

4.4/5

Average Ease of Use

37s

Avg Completion Time

4.75/5

Likelihood to Use

Task by Task Performance

Task

Check Glucose

AI Food Logging

Find Snacks

Export Data

AVG Time

1m 03s

43s

28s

12s

Success

100%

Help Requests

Key Finding

Graph too small

Universal acclaim

Positive guidance worked

Exceptionally fast

Task

Check Glucose

AI Food Logging

Find Snacks

Export Data

AVG Time

1m 03s

43s

28s

12s

Success

100%

Help Requests

Key Finding

Graph too small

Universal acclaim

Positive guidance worked

Exceptionally fast

Task

Check Glucose

AI Food Logging

Find Snacks

Export Data

AVG Time

1m 03s

43s

28s

12s

Success

100%

Help Requests

Key Finding

Graph too small

Universal acclaim

Positive guidance worked

Exceptionally fast

Task

Check Glucose

AI Food Logging

Find Snacks

Export Data

AVG Time

1m 03s

43s

28s

12s

Success

100%

Help Requests

Key Finding

Graph too small

Universal acclaim

Positive guidance worked

Exceptionally fast

The Discovery Moment

The testing session that changed everything happened on Day 2. Participant 3 (the user most similar to Priya) opened the glucose tracker, stared at the graph for 10 seconds, then looked up and said:

"I can see the line going up and down, but I have no idea what my actual reading was this morning. How am I supposed to use this?"

— Participant 3, Type 2 Diabetes, age 54

She then did something I didn't expect: she picked up her paper logbook from her purse and pointed to it. "This is what I actually use. The app just shows me a pretty graph."

Key Insight:

That moment revealed a fundamental flaw: our design visualized data beautifully but made it difficult to access the specific numbers users actually needed.

What Users Loved

AI Food Recognition

“The AI food logging feature was a game changer.”

— Siddharth

All-in-One Solution

"It's nice that everything is in one app."

— Arundhati

Data Export

"Getting all my data in one place... when my endo asks for reports."

— Apurva

Critical Issues Identified

Critical

RPN: 560

Graph Too Small / Not Zoomable

"With 4 readings per day, I need detailed trends."

Affected: 3/3 users (100%)

Critical

RPN: 294

AI Accuracy Concerns for Diverse Cuisine

"If AI doesn't recognize Indian food, I won't trust it."

Affected: 2/3 users (67%)

Critical

RPN: 392

No Support for Multiple Daily Readings View

Type 1 users specifically Type 1 diabetics test 4x daily but can't view all readings clearly

Affected: 1/3 users (33%)

Moderate

RPN: 180

No Ingredient-Based Recipe Suggestions

“I have ingredients but don't know what to make."

Affected: 2/3 users (67%)

Moderate

RPN: 150

Export Button Not Fully Actionable

"Can it go straight to WhatsApp or email?"

Affected: 1/3 users (33%)

Design Decisions based on Research

A safety-first diabetes management system

Refinement Strategy

Using FMEA (Failure Mode and Effects Analysis), I prioritized 16 usability issues by Risk Priority Number (RPN = Severity × Occurrence × Detection). I focused on the top 5 solutions that accounted for 75% of total identified risk.

Higher RPN = Higher priority for fixes

Solution 1: Enhanced Glucose Visualization

The Problem

Graph too small, no zoom, couldn't view 4+ daily readings. Type 1 users (33%) testing 4x daily abandoned feature.

Critical

RPN: 560

Design Decision

Redesign for intensive users (Type 1) as primary use case, not edge case. They make critical insulin dosing decisions multiple times daily.

Before

After

Solution

Increased graph size: 200px → 350px (75% larger)

Pinch-to-zoom gesture for detailed inspection

Multi-reading view: 4-6 daily readings with contextual labels (Fasting, Post-Breakfast, Pre-Lunch, Pre-Dinner, Bedtime)

Day/Week/Month toggle for trend analysis

Increased graph size: 200px → 350px (75% larger)

Pinch-to-zoom gesture for detailed inspection

Multi-reading view: 4-6 daily readings with contextual labels (Fasting, Post-Breakfast, Pre-Lunch, Pre-Dinner, Bedtime)

Day/Week/Month toggle for trend analysis

Impact

Task completion: 1m 03s → 30s (estimated)

Help requests: 2 → 0

Supports 4x+ daily testing for Type 1 users

Solution 2: Comprehensive Meal Log & Analysis

The Problem

Users wanted to identify which food components (carbs, protein, fat) caused glucose spikes. Couldn't isolate triggers or understand personal food responses.

Critical

RPN: 294

Design Decision

Transform meal logging from simple data entry to actionable insights tool. Help users understand why their glucose responds the way it does, not just that it changed.

Before

After

Solution

AI-Powered Logging: Camera-based food recognition with 92% accuracy for 500+ foods

Component Breakdown: Macronutrient impact displayed individually.

Smart Insights: "Your glucose spikes most after high-carb meals" with personalized patterns

Portion Detection: AI estimates serving sizes from photos (within 20% accuracy)

AI-Powered Logging: Camera-based food recognition with 92% accuracy for 500+ foods

Component Breakdown: Macronutrient impact displayed individually.

Smart Insights: "Your glucose spikes most after high-carb meals" with personalized patterns

Portion Detection: AI estimates serving sizes from photos (within 20% accuracy)

Impact

Users identify specific food triggers for glucose spikes

Enables data-driven dietary adjustments

Eliminates tedious manual entry (43s average logging time)

Solution 3: Medication Tracking & Smart Reminders

The Problem

Users managing diabetes medications needed reminders, adherence tracking, and refill alerts. 33% also needed blood pressure medication tracking for comorbidities.

High

RPN: 200

Design Decision

Medication adherence is as critical as glucose monitoring. Elevate medications to first-class feature with dedicated navigation and proactive reminders—not buried in settings.

Before

After

Solution

Medication List: Name, dosage, schedule clearly displayed with visual medication icons

Smart Reminders: Time-based notifications with quick actions (Taken/Skip/Snooze)

Adherence Tracking: 7-day calendar view showing dose compliance patterns (e.g., 13/14 doses = 93%)

Refill Alerts: 7-day advance warning before medications run out

Medication List: Name, dosage, schedule clearly displayed with visual medication icons

Smart Reminders: Time-based notifications with quick actions (Taken/Skip/Snooze)

Adherence Tracking: 7-day calendar view showing dose compliance patterns (e.g., 13/14 doses = 93%)

Refill Alerts: 7-day advance warning before medications run out

Impact

Elevated from buried feature to dedicated navigation tab

Prevents missed doses through proactive reminders

93% average adherence rate tracked

Solution 4: Streamlined Doctor Sharing

The Problem

Export completed in 12s but required 3 manual steps to share with doctors. Clinical workflow was incomplete.

Moderate

RPN: 150

Design Decision

Clinical utility requires end-to-end workflow. "Easy" means 2 taps total, not export + manual forwarding.

Before

After

Solution

Multi-format Export: PDF (doctor-friendly) + Excel/CSV (detailed analysis)

Native sharing: WhatsApp, Gmail, Messages, Print

PDF preview before sending

One-tap workflow: Generate → Preview → Share (2 taps)

Multi-format Export: PDF (doctor-friendly) + Excel/CSV (detailed analysis)

Native sharing: WhatsApp, Gmail, Messages, Print

PDF preview before sending

One-tap workflow: Generate → Preview → Share (2 taps)

Impact

Workflow: 5 steps → 2 taps

80% share directly from app

Maintains 5.0/5 satisfaction, 12s completion

v1.0 → v2.0 Complete Impact

Metric

Total Risk (RPN)

Critical Issues

Serious Issues

Task Success

Ease of Use

Safety Status

v1.0

2,046

100%

4.4 / 5

Compromised

v2.0

430

100%

4.8 / 5

Validated

Change

-79%

100% fixed

Maintained

+9%

Achieved

Metric

Total Risk (RPN)

Critical Issues

Serious Issues

Task Success

Ease of Use

Safety Status

v1.0

2,046

100%

4.4 / 5

Compromised

v2.0

430

100%

4.8 / 5

Validated

Change

-79%

100% fixed

Maintained

+9%

Achieved

Metric

Total Risk (RPN)

Critical Issues

Serious Issues

Task Success

Ease of Use

Safety Status

v1.0

2,046

100%

4.4 / 5

Compromised

v2.0

430

100%

4.8 / 5

Validated

Change

-79%

100% fixed

Maintained

+9%

Achieved

Metric

Total Risk (RPN)

Critical Issues

Serious Issues

Task Success

Ease of Use

Safety Status

v1.0

2,046

100%

4.4 / 5

Compromised

v2.0

430

100%

4.8 / 5

Validated

Change

-79%

100% fixed

Maintained

+9%

Achieved

Solution Overview: The Final Product

After 7 major refinements based on usability testing, the diabetes management app transformed from a feature-complete prototype to a product users genuinely want in their daily lives. Here's what we built.

Checkout Prototype

Feature 1: Enhanced Glucose Visualization

Multi-Reading Dashboard with Zoom View 4-6 daily glucose readings clearly. Pinch-to-zoom for detailed inspection. Day/Week/Month toggle for comprehensive trend analysis.

Key Specs:

350px graph (75% larger than v1.0)

Pinch-to-zoom gesture support

4-6 daily readings with context labels

Color-coded ranges per reading (Green/Orange/Red)

Day/Week/Month view switcher

Reduced task time: 1m 03s → 30s

"I can finally see all 4 of my daily readings clearly and understand my patterns. This changes everything."

Apurva

Enhanced Glucose Visualization

Feature 2: AI-Powered Meal Logging

AI Food Recognition with Cultural Accuracy Snap a photo, get instant nutritional breakdown. 92% accuracy for 500+ foods including comprehensive Indian cuisine (dal, paratha, dosa, rajma, biryani, sambar).

Key Specs:

Camera-based instant recognition

500+ Indian dishes in database

92% accuracy (up from 80%)

35 second average logging time

AI portion size detection (within 20%)

Component glucose impact prediction

"The AI recognized my dal and paratha perfectly. This is the feature that makes me switch apps."

Siddharth

AI Powered Meal Logging and Recommendation

Feature 3: Smart Medication Management

Never Miss a Dose Track diabetes + comorbidity medications with smart reminders. View 7-day adherence calendar. Share compliance data with doctor.

Key Specs:

Medication list with dosage schedules

Time-based smart reminders

Quick actions from notification (Taken/Skip/Snooze)

7-day adherence calendar (93% average)

7-day advance refill alerts

Included in doctor export reports

"The quick 'Taken' button is perfect. I can see my adherence and share it with my doctor."

Arundhati

Smart Medication Management

Feature 4: Streamlined Doctor Sharing

2-Tap Doctor Collaboration Generate comprehensive reports in 8 seconds. Preview PDF. Share directly via WhatsApp, email, or print. No manual forwarding needed.

Key Specs:

Multi-format: PDF (doctor-friendly) + Excel/CSV

Native sharing: WhatsApp, Gmail, Messages, Print

PDF preview before sending

8 second average completion

Includes glucose, meals, medications, adherence

5.0/5 satisfaction maintained

"Getting all my data in one place when my endo asks for reports—no more compiling manually."

Apurva

Streamlined Doctor Sharing

Outcome & Impact

100%

Task Success Rate Maintained with zero help requests

4.9 / 5

Overall Satisfaction

Up from 4.78/5 (+4%)

-32%

Faster Completion

37s → 25s average

-79%

Risk Reduction

2,046 → 430 RPN

v1.0 → v2.0 Complete Impact

Metric

Total Risk (RPN)

Critical Issues

Serious Issues

Task Success

Ease of Use

Safety Status

v1.0

2,046

100%

4.4 / 5

Compromised

v2.0

430

100%

4.8 / 5

Validated

Change

-79%

100% fixed

Maintained

+9%

Achieved

Metric

Total Risk (RPN)

Critical Issues

Serious Issues

Task Success

Ease of Use

Safety Status

v1.0

2,046

100%

4.4 / 5

Compromised

v2.0

430

100%

4.8 / 5

Validated

Change

-79%

100% fixed

Maintained

+9%

Achieved

Metric

Total Risk (RPN)

Critical Issues

Serious Issues

Task Success

Ease of Use

Safety Status

v1.0

2,046

100%

4.4 / 5

Compromised

v2.0

430

100%

4.8 / 5

Validated

Change

-79%

100% fixed

Maintained

+9%

Achieved

User Feedbacks

Finally Understand My Food

"I tested it with dal, paratha, and dosa—it recognized everything perfectly. Now I trust it enough to use it for every meal. This is the feature that makes me switch apps."

— Siddharth (Type 2, 3 years)

Everything in One Place

"Glucose tracking, meal logging, medication reminders, doctor reports all in one app. I can finally delete MySugr, Healthify Me, and my medication reminder app."

— Arundhati (Type 2, 5 years)

Now I Can Actually Use This

"The zoom feature is exactly what I needed. I can finally see all 4 of my daily readings clearly and understand my patterns. This changes everything for managing my insulin."

— Apurva (Type 1, 16 years)

COMPETITIVE ADVANTAGE

Feature

AI Recognition

Diverse Cuisine Analysis

Multi-Reading Graph

Medications

Ingredient Recipes

User Satisfaction

Diabetes:M

Manual

Limited

Basic

Manual

3.8 / 5

mySugr

Manual

Limited

Basic

None

3.8 / 5

Glucose Buddy

Manual

Limited

Basic

Smart

None

3.5 / 5

Our solution

92%

500 + Recipies

Advanced Analysis

Smart Tracking

Proactive

4.91 / 5

Feature

AI Recognition

Diverse Cuisine Analysis

Multi-Reading Graph

Medications

Ingredient Recipes

User Satisfaction

Diabetes:M

Manual

Limited

Basic

Manual

3.8 / 5

mySugr

Manual

Limited

Basic

None

3.8 / 5

Glucose Buddy

Manual

Limited

Basic

Smart

None

3.5 / 5

Our solution

92%

500 + Recipies

Advanced Analysis

Smart Tracking

Proactive

4.91 / 5

Feature

AI Recognition

Diverse Cuisine Analysis

Multi-Reading Graph

Medications

Ingredient Recipes

User Satisfaction

Diabetes:M

Manual

Limited

Basic

Manual

3.8 / 5

mySugr

Manual

Limited

Basic

None

3.8 / 5

Glucose Buddy

Manual

Limited

Basic

Smart

None

3.5 / 5

Our solution

92%

500 + Recipies

Advanced Analysis

Smart Tracking

Proactive

4.91 / 5

Feature

AI Recognition

Diverse Cuisine Analysis

Multi-Reading Graph

Medications

Ingredient Recipes

User Satisfaction

Diabetes:M

Manual

Limited

Basic

Manual

3.8 / 5

mySugr

Manual

Limited

Basic

None

3.8 / 5

Glucose Buddy

Manual

Limited

Basic

Smart

None

3.5 / 5

Our solution

92%

500 + Recipies

Advanced Analysis

Smart Tracking

Proactive

4.91 / 5

AI in UX Design: What Worked, What Didn't

Evaluating the role of Artificial Intelligence in the design process

Dovetail AI

Thematic analysis

Figma Make + Claude

Prototype generation

Figma AI

PRD compliance check

AI-Assisted FMEA

Risk scoring

Phase

Research

Design

Testing

Analysis

Tool

Dovetail AI

Figma Make

Claude

Figma MCP + Claude

AI - FEMA

Usage

Processed 3 user interviews

Generated initial UI components

Content generation for onboarding

Automated heuristics check

Risk probability analysis

Outcome

34 FDA requirements identified

18 screens, 62% requirement coverage

Persona Development, alert messaging

31 issues found, prevented testing with critical gaps

16 usability issues prioritized by PRN

Phase

Research

Design

Testing

Analysis

Tool

Dovetail AI

Figma Make

Claude

Figma MCP + Claude

AI - FEMA

Usage

Processed 3 user interviews

Generated initial UI components

Content generation for onboarding

Automated heuristics check

Risk probability analysis

Outcome

34 FDA requirements identified

18 screens, 62% requirement coverage

Persona Development, alert messaging

31 issues found, prevented testing with critical gaps

16 usability issues prioritized by PRN

Phase

Research

Design

Testing

Analysis

Tool

Dovetail AI

Figma Make

Claude

Figma MCP + Claude

AI - FEMA

Usage

Processed 3 user interviews

Generated initial UI components

Content generation for onboarding

Automated heuristics check

Risk probability analysis

Outcome

34 FDA requirements identified

18 screens, 62% requirement coverage

Persona Development, alert messaging

31 issues found, prevented testing with critical gaps

16 usability issues prioritized by PRN

Phase

Research

Design

Testing

Analysis

Tool

Dovetail AI

Figma Make

Claude

Figma MCP + Claude

AI - FEMA

Usage

Processed 3 user interviews

Generated initial UI components

Content generation for onboarding

Automated heuristics check

Risk probability analysis

Outcome

34 FDA requirements identified

18 screens, 62% requirement coverage

Persona Development, alert messaging

31 issues found, prevented testing with critical gaps

16 usability issues prioritized by PRN

What AI Did Well

Using AI for Rapid Iteration

Figma Make let me test 18 screens quickly

Identified broken navigation links

Verified form field labels

AI Pre-Testing Before Human Testing

AI found 31 issues, fixed 14 critical ones

Prevented testing with broken accessibility

Made human testing more productive (focused on UX, not bugs)

Human Testing for Real-World Validation

3 participants revealed 5 issues AI couldn't predict

User quotes led to specific design fixes

Cultural needs emerged (Indian cuisine accuracy)

What AI Missed

Over-Relying on AI Assessment

AI said 93% = "ready to launch"

Reality: Still had 5 critical usability issues

Should have been skeptical of perfect AI scores

Not Testing for Edge Cases Earlier

Apurva tests 4x daily (Type 1)

Designed for 1-2x daily users (Type 2)

Should have recruited Type 1 users from the start

Assuming AI Understood Context

AI validated meal logging "worked"

Didn't understand users wanted recipes, not just logs

Cultural food database gaps invisible to AI

AI vs. Human Testing

What was tested

Features Work

Meet PRD

Accessibility

Graph Usability

AI Food Recognition

Meal Planning

Export Feature

AI Evaluation (93% Grade A)

All features functional

93% compliance

WCAG 2.1 AA met (after fixes)

Graph renders correctly

Camera works, recognizes food

Meal logging works

PDF generation works

Human testing (91.7% Score)

100% task success rate

Validated all requirements

Font size 4.3/5, readability good

Too small for 4x daily readings (4.0 / 5)

Needs diverse cuisine accuracy (67% concerned)

Users want proactive recipes, not just logging

Needs direct sharing, not just save (33%)

What was tested

Features Work

Meet PRD

Accessibility

Graph Usability

AI Food Recognition

Meal Planning

Export Feature

AI Evaluation (93% Grade A)

All features functional

93% compliance

WCAG 2.1 AA met (after fixes)

Graph renders correctly

Camera works, recognizes food

Meal logging works

PDF generation works

Human testing (91.7% Score)

100% task success rate

Validated all requirements

Font size 4.3/5, readability good

Too small for 4x daily readings (4.0 / 5)

Needs diverse cuisine accuracy (67% concerned)

Users want proactive recipes, not just logging

Needs direct sharing, not just save (33%)

What was tested

Features Work

Meet PRD

Accessibility

Graph Usability

AI Food Recognition

Meal Planning

Export Feature

AI Evaluation (93% Grade A)

All features functional

93% compliance

WCAG 2.1 AA met (after fixes)

Graph renders correctly

Camera works, recognizes food

Meal logging works

PDF generation works

Human testing (91.7% Score)

100% task success rate

Validated all requirements

Font size 4.3/5, readability good

Too small for 4x daily readings (4.0 / 5)

Needs diverse cuisine accuracy (67% concerned)

Users want proactive recipes, not just logging

Needs direct sharing, not just save (33%)

What was tested

Features Work

Meet PRD

Accessibility

Graph Usability

AI Food Recognition

Meal Planning

Export Feature

AI Evaluation (93% Grade A)

All features functional

93% compliance

WCAG 2.1 AA met (after fixes)

Graph renders correctly

Camera works, recognizes food

Meal logging works

PDF generation works

Human testing (91.7% Score)

100% task success rate

Validated all requirements

Font size 4.3/5, readability good

Too small for 4x daily readings (4.0 / 5)

Needs diverse cuisine accuracy (67% concerned)

Users want proactive recipes, not just logging

Needs direct sharing, not just save (33%)

What I learned about AI in UX

AI Excels at Technical Validation

Found 31 technical issues, validated 93% PRD compliance, and caught accessibility gaps—but couldn't predict real user behavior patterns or usage contexts.

AI Saved Time, Not Critical Insights

Generated 18 screens (~20 hours saved), auto-tagged 247 insights (~4 hours saved), found 31 issues (~8 hours saved). But 3 hours of human testing found 5 issues that mattered more for product success.

AI Cannot Assess Mental Models

Confirmed all features existed and worked, but couldn't predict where users expected to find them. Missed that 100% needed help finding morning summary and graph was too small for intensive users.

AI + Human = Necessary Combination

AI caught 14 critical technical issues. Humans caught 5 critical usability issues. Without AI: broken accessibility in testing. Without humans: unusable graphs at launch. Both needed, neither sufficient.

AI Excels at Technical Validation

Found 31 technical issues, validated 93% PRD compliance, and caught accessibility gaps—but couldn't predict real user behavior patterns or usage contexts.

AI Cannot Assess Mental Models

Confirmed all features existed and worked, but couldn't predict where users expected to find them. Missed that 100% needed help finding morning summary and graph was too small for intensive users.

AI Saved Time, Not Critical Insights

AI + Human = Necessary Combination

AI Excels at Technical Validation

Found 31 technical issues, validated 93% PRD compliance, and caught accessibility gaps—but couldn't predict real user behavior patterns or usage contexts.

AI Cannot Assess Mental Models

Confirmed all features existed and worked, but couldn't predict where users expected to find them. Missed that 100% needed help finding morning summary and graph was too small for intensive users.

AI Saved Time, Not Critical Insights

AI + Human = Necessary Combination

What This Project Taught Me

This project fundamentally changed how I think about design validation. I learned that AI is a powerful accelerator, but human insight is irreplaceable.

The AI heuristic evaluation gave me a 93% "Grade A" and told me the prototype was ready to launch. I believed it. Then three hours of human testing revealed five critical issues that would have caused real-world failures—a graph too small for intensive users, missing cultural food accuracy, and workflow gaps AI never detected.

The turning point: When Apurva said "I test 4 times a day, I need to see detailed trends," I realized I had designed for average users, not the people who needed the app most. AI validated the graph "worked." Apurva showed me it was unusable for her survival.

Key lessons learned:

Good design isn't enough for healthcare—safe design is the baseline.

The gap between 93% technical validation and 2 critical safety issues taught me that compliance ≠ usability.

Design for edge cases, not averages.

Type 1 diabetics testing 4x daily aren't edge cases—they're the users who need the app most. Designing for them improved the experience for everyone.

Prioritize by risk, not effort.

Some 10-minute fixes (graph zoom) had RPN 560. Some 10-hour features had RPN 50. FMEA taught me where to focus energy.

Zero regression matters.

All five features with perfect 5.0/5 ratings (AI food logging, data export, navigation, medications) remained untouched during refinements. Fixing critical issues while preserving strengths is the real challenge.

Good design isn't enough for healthcare—safe design is the baseline.

The gap between 93% technical validation and 2 critical safety issues taught me that compliance ≠ usability.

Prioritize by risk, not effort.

Some 10-minute fixes (graph zoom) had RPN 560. Some 10-hour features had RPN 50. FMEA taught me where to focus energy.

Design for edge cases, not averages.

Type 1 diabetics testing 4x daily aren't edge cases—they're the users who need the app most. Designing for them improved the experience for everyone.

Zero regression matters.

Good design isn't enough for healthcare—safe design is the baseline.

The gap between 93% technical validation and 2 critical safety issues taught me that compliance ≠ usability.

Prioritize by risk, not effort.

Some 10-minute fixes (graph zoom) had RPN 560. Some 10-hour features had RPN 50. FMEA taught me where to focus energy.

Design for edge cases, not averages.

Type 1 diabetics testing 4x daily aren't edge cases—they're the users who need the app most. Designing for them improved the experience for everyone.

Zero regression matters.

What I'd Do Differently

Test Stress Scenarios Earlier

Simulate panic/hypoglycemia conditions during initial concept testing, not just at the end.

Conduct Findability Testing

Run dedicated tree tests to ensure critical features (Log Insulin) are discoverable in under 5 seconds.

More Diverse Participants

Include users with visual impairments and motor control issues earlier in the validation phase.

Future Roadmap

Phase 2

Months 1-3

Wearable integration (Apple Health)

Carb counting camera (AI)

Caregiver dashboard mode

Dark mode for night use

Phase 3

Months 4-6

Telehealth video visits

Prescription refill integration

Community support forums

Advanced trend analytics

Phase 4

Months 7-12

Automated insulin pump loops

Voice command logging

Predictive glucose alerts

Multi-language support

Personal Growth

Designing for healthcare isn't just about solving problems—it's about honoring the vulnerability of the user. This project taught me that in safety-critical domains, safe design is the baseline, not a feature. Moving forward, I carry the responsibility that every interface decision has real-world consequences, and that true empathy requires looking beyond the "happy path" to protect users when they are at their most vulnerable.

Thank you for your curiosity

Connect with me for collaboration, coffee, or fun.