Fluency Medical

By integrating large language models and natural language processing, we've developed features to streamline physicians' administrative tasks.
Duration
Role
Team
36 Hours
• Project Management
• UX Design
• UI Design
• Visual Design
• 1 designer
• 3 engineers

Background

When I worked in healthcare administration, I was able to understand the inner workings of the healthcare system and just how complicated it was. This project started at a hackathon, with 36 hours for start and completion, where my team and I had to create a product incorporating large language models and natural language processing. Through multiple time constraints, we only created the backbone of the product, but now are on the journey of making it a fully working product.

Stay tuned till the end for more information about our success :)

Problem Space

Physicians are often burdened with administrative tasks because the complexity of healthcare documentation, insurance coding, and the need for clinical expertise make it challenging for administrative assistants to handle these responsibilities. In the American healthcare system, precise diagnosis codes and paperwork are essential for insurance reimbursement–for each patient appointment– and this requires the direct involvement of the treating physician to ensure accurate documentation and fair compensation.

In other words, since the doctor creates the diagnosis, they are the ones that know the corresponding billing codes.

Now, using the tools we have in the next 36 hours, how might we help them reduce these tedious daily tasks?

Our Solution

Centered on telehealth platforms, the web application is designed to take off some unnecessary cognitive load for physicians.
1. The patient inputs their symptoms into the GPT chat bot from the patient portal which send an analysis and summary to the physician’s end.
2. The physician’s end would receive possible diagnosis and next actions.
3. Physicians would approve and select the recommended action(s), sending the action request to appropriate facilities and billing codes to corresponding insurance companies

Process

Research

Ideation

Designs

• Usability testing
• Key takeaways
• User flow
• Sketches
• Low Fidelity Wireframes
• High Fidelity Wireframes
• Visual Design

Research

Surveys were created and sent out to my surrounding community of healthcare worker friends. The problem space that I ideated was further confirmed and validated through their own experiences
Key Takeaways:
1. Patient notes were always created through the process of multitasking, unless they had the costs to hire a medical scribe. Most of the time, patient notes would have to be finished after the workday.
2. Insurance billing was tedious and inefficient, where each appointment, no matter which patient had to have an insurance bill made in order for the physician to get paid.

So, how might we use large language models and natural language processing to make insurance billing, diagnosis and medical transcription less time consuming?

Ideation

As I created a user flow, I opted for he matrix structure, keeping all tabs and screens accessible.
While creating sketches, I wanted to keep everything accessible and familiar for the user.

Design

Low Fidelity
As I started the low fidelity, easy accessibility was always in the front of my mind. When designing for such busy users, they have to be able to clearly understand all icons and features. In addition, we have to keep in mind their familiarity bias and how it would hinder or benefit them during their busy schedule.
High Fidelity Wireframes
Visual Design

What I Learned

Effective planning played a pivotal role in our project's outstanding outcome. Despite facing time constraints, I allocated sufficient time to thoroughly research healthcare processes, resulting in well-structured surveys and interviews that yielded invaluable insights. This approach not only facilitated a streamlined development process for our team but also enhanced my proficiency in HTML, CSS, and Javascript.

Further Background

My teammates and I entered CalHacks, Berkley's 36 hour AI hackathon and created Fluency Medical. Among the 800 project submissions, we secured a spot in the esteemed top 50, giving us an opportunity to have lunch with a venture capital firm, SkyDeck. It was truly an honor to have received recognition from distinguished investors, including those from renowned AI companies such as Llama and Roger AI. Their positive feedback and enthusiasm have served as a testament to the value and potential impact of Fluency Medical in the market.
Pictured Left: My teammate, Mayolo and I presenting our product to an academic from Oxford.
Pictured Right: My CalHack teammates and developer mentor (from left to right) Mayolo Valencia, Homen Shum, Vincent Tseng, William Redenbaugh and me!

View the Prototype

View on Figma

Let's Connect!

Instagram Logo
Mail icon