Kymari Bratton

Healthcare AI

Brightside Health

A collaborative AI project exploring how a knowledge graph interface could make mental health information easier to interpret and navigate.

I served as Front End Lead and also contributed as an AI Engineer, focusing most of my time on improving relationships within the graph experience and shaping how the product would actually feel to use.

View project repo Back to projects

Collaborative concept and engineering exploration.

Brightside Health knowledge graph screens

Overview

Making complexity easier to read

The project explored how a knowledge graph experience could surface complex mental health relationships without losing the nuance that makes the data useful. My role centered on the interface layer: structure, readability, interaction, and how people could move through the graph without feeling lost.

Project details

Type

Collaborative AI project

Role

Front End Lead · AI Engineer

Focus

Relationship improvement · Graph UX

Tools

React · Cytoscape.js · Neo4j · Figma

Problem

Dense information needs stronger UX.

The problem

Mental health information is technical, interconnected, and often hard to parse. The challenge was figuring out how a graph interface could make those relationships easier to search, scan, and understand without oversimplifying them.

Why it mattered

Knowledge graphs are only useful if people can interpret them. Even strong backend work starts to lose value when the front end feels cluttered, confusing, or too technical to confidently navigate.

86.0% Best reported model-based accuracy in the project documentation.
81.81% Reported precision score for extracted relationships in the repo notes.

Research

What informed the direction

Intended users

The experience was designed with clinicians in mind, but one of the biggest lessons from the project was that assumptions are not the same thing as direct research.

  • Manual annotation
  • Team collaboration
  • Technical feedback
  • UI exploration

My research reality

At the time I was still early in my UX learning, so a lot of the design choices came from what I thought clinicians would appreciate rather than from enough interviews or usability testing. That gap became one of the clearest takeaways.

01 Relationship data gets difficult to interpret fast without clear hierarchy.
02 Search and filtering are essential when the graph begins to sprawl.
03 Too much information can make a graph less useful instead of more helpful.
04 Strong UX is what helps technical work become something people can actually use.

Direction

Designing the graph around comprehension.

North star

Make a technical graph experience easier to explore, easier to read, and more useful for the people trying to interpret complex relationships.

  • Reduce visual overload
  • Support focused exploration
  • Clarify relationships
  • Improve front-end usability

What I worked on

  • Manual relationship extraction. I helped ground the experience in clearer examples of how entities and connections should appear.
  • UI model exploration. I explored graph interface directions that made reading and navigating feel less overwhelming.
  • Front-end UX/UI leadership. I focused on how the product should feel from a usability and interaction standpoint.

Final concept

  • Upload and retrieve graphs. Support ongoing analysis rather than one-time viewing.
  • Search-first exploration. Help users find relevant nodes and relationships faster.
  • Dynamic visualization. Use visual cues to make relationships easier to read.
  • Custom filtering. Let users reduce clutter and narrow focus when needed.

Impact

What the project produced

  • A front-end interface direction for exploring mental health knowledge graphs.
  • A shared workflow across UI, graph integration, evaluation, and documentation.
  • A stronger foundation for improving relationship quality with future LLM work.

Reflection

What I learned

This project taught me how much UX matters in technical spaces. I also got more hands-on experience with Neo4j, Jupyter Notebook, and how backend structure connects to front-end decisions.

If I continued the work, I would simplify the information architecture, strengthen accessibility, and bring in more direct clinician research before pushing the interface further.