EdenCare - AI Analytics POC
Eden Care works with large volumes of medical, operational, and engagement data — from clinic visits and claims to HR records and wellness activity
Internal teams needed insights from this data, but most staff:
Didn’t have SQL or technical skills
Found dashboards slow and hard to use
Spent hours or days preparing reports
The AI team wanted to explore a new idea:
Could staff simply ask questions in plain English and get useful insights instantly?
This project was a short proof of concept, designed to help the company decide whether this idea was worth deeper investment.
This experiment focused on answering a small set of critical questions:
Would non-technical staff understand how to ask questions in natural language?
Would they trust AI-generated results enough to use them in real work?
Could this approach meaningfully reduce reporting time?
The goal was not to build a full analytics product — only to gather enough signal to decide whether to continue.
One-week timeline
Needed something testable, not complete
Had to be simple enough for stakeholders to evaluate quickly
These constraints shaped every design decision.
I designed the experiment end-to-end.
I was responsible for
Shaping the validation scope
Designing the minimum interface needed to test the idea
Making AI output understandable and trustworthy
Working closely with the AI and engineering teams to ensure feasibility
The focus was speed, clarity, and learning — not polish.
I intentionally kept the flow small and focused on three things that mattered for validation. Asking questions, understanding the result and creating trust and transparency regarding how the result was generated.
The Experiment
Staff could type questions in plain English or select from example prompts.
The goal was to see whether people could ask meaningful questions without training.
Trust was a major risk in this experiment, especially in a healthcare context.
To address this, I designed a transparency panel that showed
How the AI interpreted the question
The steps taken to generate the answer
The underlying database query in MonGo DB
This helped stakeholders feel confident the system wasn’t a “black box” and also helped the AI team debug early outputs.
Results were shown in three simple formats:
A table for raw data
A chart chosen automatically based on the dataset
A short written summary explaining what the data meant
This made insights easy to understand without dashboards or SQL
The product is still in active development. This case study reflects real design decisions made during build, not a polished post-launch story.
Early validation showed strong potential:
~70% of stakeholders responded positively and approved the concept for further development
A workflow that typically took ~3 days (SQL + reporting) could be reduced to ~5 minutes
Non-technical staff could get insights without relying on engineering teams
The transparency view significantly increased trust in AI outputs
This experiment gave the team enough confidence to decide whether to proceed with a full build.
Experiments work best when they stay small and focused
Trust is just as important as accuracy in AI-driven tools
Showing how AI arrives at answers increases confidence and adoption
The biggest takeaway for me was learning to strip ideas down to the smallest version that clearly communicates value.
I included this project to show how I approach early-stage validation, AI-powered products and designing just enough to support decision making. It reflects my ability to move quickly, design with intent and avoid overbuilding before direction is confirmed.
Up next
