Epic Systems

Epic Systems

Zero -> AI design system used by 100K+ clinicians and 350+ orgs.

Enterprise AI

B2B

Securing Pilots

Disclaimer: Due to IP restrictions, design artifacts cannot be shared.

Epic, the largest EHR provider in the U.S., needed a way to make AI safe, consistent, and adoption-ready across 80+ apps.

I co-built Epic’s first AI design system from scratch, unifying scattered AI efforts into a trusted foundation now used by 80K+ clinicians and 330+ organizations.

The Results

The system became Epic’s foundation for all AI experiences:

  • Used by 100K+ clinicians across 350+ orgs

  • Launched for Epic’s largest annual conference (40K+ attendees)

  • Endorsed by Epic’s C-suite as a “beautiful, scalable solution”

  • Established validated standards still guiding future AI work

"Working with Josh was a fantastic experience. He was highly responsive to feedback and played a key role in shaping our ideas into a compelling narrative and an enhanced user experience around AI.

I especially appreciated his initiative in researching the best solutions and his creativity in exploring new directions."

Nick Marzotto

Head of AI Customer Success, Epic

The Process

Step 1

Define the foundation

Audited 30+ in-progress AI features to uncover adoption blockers: lack of trust, unclear ownership, and no shared UX standards.

Step 1

Define the foundation

Audited 30+ in-progress AI features to uncover adoption blockers: lack of trust, unclear ownership, and no shared UX standards.

Step 2

Establish visual identity

Created a consistent AI identity with a new icon system, visual language, and the “Bloom” symbol as the universal AI marker.

Step 2

Establish visual identity

Created a consistent AI identity with a new icon system, visual language, and the “Bloom” symbol as the universal AI marker.

Step 3

Create reusable patterns

Defined four core workflows (summarization, automation, content transformation, drafted text) and built reusable components.

Step 3

Create reusable patterns

Defined four core workflows (summarization, automation, content transformation, drafted text) and built reusable components.

Step 4

Scale and validate

Rolled out across 80+ apps and tested continuously with clinicians to ensure safety, clarity, and adoption.

Step 4

Scale and validate

Rolled out across 80+ apps and tested continuously with clinicians to ensure safety, clarity, and adoption.

Solving Hard Problems at Scale

These were some of the UX problems solved, where design directly impacted trust, adoption, and safety.

The AI design system

Before: Scattered teams, inconsistent layouts, lost ideas. After: A unified framework and shared AI language used across 80+ apps.

The AI design system

Before: Scattered teams, inconsistent layouts, lost ideas. After: A unified framework and shared AI language used across 80+ apps.

Trust and explainability

Before: AI outputs lacked context or reliability. After: Designed patterns for confidence scores, transparency, and control, so clinicians could trust and act confidently.

Trust and explainability

Before: AI outputs lacked context or reliability. After: Designed patterns for confidence scores, transparency, and control, so clinicians could trust and act confidently.

Visual identity + brand

Before: AI had no recognizable brand or “face.” After: Co-designed a enterprise brand and style system for AI that made outputs instantly recognizable and interpretable.

Visual identity + brand

Before: AI had no recognizable brand or “face.” After: Co-designed a enterprise brand and style system for AI that made outputs instantly recognizable and interpretable.

Scalability

Before: Every AI feature was a one-off pilot. After: Shifted from “features” to “systems,” enabling faster, safer scale with trust built in.

Scalability

Before: Every AI feature was a one-off pilot. After: Shifted from “features” to “systems,” enabling faster, safer scale with trust built in.

Broader Scope of Work

Beyond design, I helped define Epic’s broader AI direction:

Partnered with leadership to shape AI strategy

Built pitch decks and roadmaps with partners (e.g. Microsoft, Mayo Clinic)

Advised 80+ internal teams as an “AI design consultant”

Delivered executive presentations to evangelize AI design

Built design systems, Figma component libraries, and internal guidelines

Defined future vision roadmaps for AI’s future in healthcare

This role blended strategy and execution, driving alignment across teams and leadership.

Outcomes and Impact

The unified AI design system reshaped how 80+ clinical apps delivered AI. Teams could now scale safely through shared components, consistent identity, and built-in safeguards, reducing duplication and accelerating development.

The system influenced Epic’s broader AI strategy, partner presentations, and long-term vision, proving that strong UX can drive both adoption and enterprise direction at scale.

Lessons for Founders

The same patterns apply to AI startups building their first or next product:

Think in systems, not features: Reusable patterns and safeguards enable scale from 1 to 100.

Build trust into the workflow: Explainability and user control are levers for strong adoption.

Position for impact: A strong design system builds credibility, partnerships, and strategic clarity.

This project showed how UX can turn fragmented AI pilots into systems users trust—setting the foundation for scalable, safe AI adoption.

← Back to home

Managing diabetes is exhausting. Every meal means constant math: counting carbs, calculating insulin, remembering to administer insulin every meal. One slip can throw everything off.

I built Steady after seeing my girlfriend, Kiara, face that challenge daily with type 1 diabetes. It’s an AI-powered MVP that makes tracking faster, easier, and less stressful.

Project overview

Steady is a concept for an AI powered carb and insulin tracking app that helps people with diabetes understand how specific foods affect their glucose.

  • Connects food logging with real time CGM data

  • Shows how meals impact glucose trends

  • Helps users make faster, more confident food decisions

This project started as a personal side project inspired by my girlfriend Kiara, who manages type 1 diabetes. The first version explored the problem space and resulted in an early prototype.

I gathered feedback from registered dietitians and diabetes specialists at the University of Michigan along with early user feedback. This project represents a rebuilt V2 based on those insights.

Latest designs

Methodology

Research and organization

  • Consolidated prior research from the first Steady prototype

  • Synthesized interviews with University of Michigan clinicians and dietitians

  • Included direct conversations with people managing diabetes

  • Used Perplexity for desk research on CGM workflows, diabetes management, and nutrition tracking

  • Organized insights and patterns in one place to clarify the problem space

Define the product

  • Used Figma and FigJam to define the problem statement and primary persona

  • Mapped current diabetes food and glucose workflows

  • Designed an optimized workflow for connecting meals and glucose data

  • Created the early information hierarchy to structure the product

  • Gathered interface ideas and references based on this structure

AI-assisted ideation

  • Used the research and product structure to build a foundational prompt

  • Tested early concept generation using Lovable, Google Stitch, and Figma Make

  • Evaluated outputs across tools as an ideation method

  • Consolidated the strongest design direction into Lovable

Rapid design and prototyping

  • Designed the product, prototype, and early brand directly in Lovable

  • Iterated on UX and functionality simultaneously through rapid prototyping

  • Gathered continuous feedback from my girlfriend who manages Type 1 diabetes

  • Refined the product through multiple iterations over several days

Note: Version 2 of Steady is still in progress.

Results and impact

The MVP gave Kiara a faster, more supportive flow, without mental math. She can log meals, get AI carb and insulin estimates, and track everything in one place.

Clinicians who reviewed the prototype called it a clear improvement over existing tools, with one stating they’d recommend Steady over GluRoo as the “superior” option for patients.

"If this becomes operational, I would be more than happy to recommend this rather than GluRoo [leading competitor] to our patients.

It would give the value that the diabetes population needs - so this would make it superior."

Katy G.

Registered Dietitian; Certified Diabetes Specialist - University of Michigan Health

V1: Designs and live prototype

Try the live prototype of the first version of Steady: Open the live Steady MVP

Please note that this is the earlier version of the MVP and is no longer being actively worked on. The design and brand system for Steady is reflective in the 'Latest Designs' section above.

How to explore (takes 30 seconds):

  1. Enter the app and click (+)

  2. Tap Upload Meal to add a photo (or use the sample images).

  3. See instant carb + insulin estimates.

  4. Navigate to History to view logged meals.

  5. Explore other tabs and features!

Important: This is an MVP demo, not medical advice. There may be bugs or placeholder content. For feedback or questions, reach out to josh@buildwithfloat.com

What's next

  • More testing with a wider set of users

  • Exploring integration with continuous glucose monitors (CGMs)

  • Shaping Steady into a real product, with stronger privacy, onboarding, and personalization built in.

We have a working MVP that feels real, so the next step is pressure-testing it in daily life and expanding its reach.

Lessons for founders

A small, focused MVP is enough to test a problem meaningfully.

In health, trust and tone carry as much weight as the feature set.

Building for a real person surfaces more insights than assumptions and personas.

This project shows how quickly a focused MVP can move from an idea to something real in a person’s hands, proving impact before it ever needs to scale.

← Back to home

Managing diabetes is exhausting. Every meal means constant math: counting carbs, calculating insulin, remembering to administer insulin every meal. One slip can throw everything off.

I built Steady after seeing my girlfriend, Kiara, face that challenge daily with type 1 diabetes. It’s an AI-powered MVP that makes tracking faster, easier, and less stressful.

Project overview

Steady is a concept for an AI powered carb and insulin tracking app that helps people with diabetes understand how specific foods affect their glucose.

  • Connects food logging with real time CGM data

  • Shows how meals impact glucose trends

  • Helps users make faster, more confident food decisions

This project started as a personal side project inspired by my girlfriend Kiara, who manages type 1 diabetes. The first version explored the problem space and resulted in an early prototype.

I gathered feedback from registered dietitians and diabetes specialists at the University of Michigan along with early user feedback. This project represents a rebuilt V2 based on those insights.

Latest designs

Methodology

Research and organization

  • Consolidated prior research from the first Steady prototype

  • Synthesized interviews with University of Michigan clinicians and dietitians

  • Included direct conversations with people managing diabetes

  • Used Perplexity for desk research on CGM workflows, diabetes management, and nutrition tracking

  • Organized insights and patterns in one place to clarify the problem space

Define the product

  • Used Figma and FigJam to define the problem statement and primary persona

  • Mapped current diabetes food and glucose workflows

  • Designed an optimized workflow for connecting meals and glucose data

  • Created the early information hierarchy to structure the product

  • Gathered interface ideas and references based on this structure

AI-assisted ideation

  • Used the research and product structure to build a foundational prompt

  • Tested early concept generation using Lovable, Google Stitch, and Figma Make

  • Evaluated outputs across tools as an ideation method

  • Consolidated the strongest design direction into Lovable

Rapid design and prototyping

  • Designed the product, prototype, and early brand directly in Lovable

  • Iterated on UX and functionality simultaneously through rapid prototyping

  • Gathered continuous feedback from my girlfriend who manages Type 1 diabetes

  • Refined the product through multiple iterations over several days

Note: Version 2 of Steady is still in progress.

Results and impact

The MVP gave Kiara a faster, more supportive flow, without mental math. She can log meals, get AI carb and insulin estimates, and track everything in one place.

Clinicians who reviewed the prototype called it a clear improvement over existing tools, with one stating they’d recommend Steady over GluRoo as the “superior” option for patients.

"If this becomes operational, I would be more than happy to recommend this rather than GluRoo [leading competitor] to our patients.

It would give the value that the diabetes population needs - so this would make it superior."

Katy G.

Registered Dietitian; Certified Diabetes Specialist - University of Michigan Health

V1: Designs and live prototype

Try the live prototype of the first version of Steady: Open the live Steady MVP

Please note that this is the earlier version of the MVP and is no longer being actively worked on. The design and brand system for Steady is reflective in the 'Latest Designs' section above.

How to explore (takes 30 seconds):

  1. Enter the app and click (+)

  2. Tap Upload Meal to add a photo (or use the sample images).

  3. See instant carb + insulin estimates.

  4. Navigate to History to view logged meals.

  5. Explore other tabs and features!

Important: This is an MVP demo, not medical advice. There may be bugs or placeholder content. For feedback or questions, reach out to josh@buildwithfloat.com

What's next

  • More testing with a wider set of users

  • Exploring integration with continuous glucose monitors (CGMs)

  • Shaping Steady into a real product, with stronger privacy, onboarding, and personalization built in.

We have a working MVP that feels real, so the next step is pressure-testing it in daily life and expanding its reach.

Lessons for founders

A small, focused MVP is enough to test a problem meaningfully.

In health, trust and tone carry as much weight as the feature set.

Building for a real person surfaces more insights than assumptions and personas.

This project shows how quickly a focused MVP can move from an idea to something real in a person’s hands, proving impact before it ever needs to scale.

← Back to home

Your turn.

Stop waiting for the perfect moment. Your MVP is 10 days away.

Build My MVP

Your turn.

Stop waiting for the perfect moment. Your MVP is 10 days away.

Build My MVP