EdenCare
Asking questions of data without writing queries
I explored natural-language analytics and inspectable AI output in a one-week freelance proof of concept.
A one-week freelance proof of concept exploring natural-language analytics and inspectable AI output.
My role & scope
I worked as a freelance product designer with the AI team, engineering and internal stakeholders.
Freelance Product Designer · One week. Collaborators: AI team, Engineering, internal stakeholders.
The problem & why it mattered
A one-week freelance proof of concept explored whether staff could ask useful questions of complex data in plain language.
EdenCare’s internal teams needed answers from complex medical and operational data. Extracting an answer required technical help and reporting work.
I designed a one-week proof of concept to explore whether staff could ask useful questions in plain language, understand the results, and inspect how an answer was produced.
Design strategy & key decisions
I scoped the experiment around asking a question, evaluating the answer and inspecting its interpretation.
With one week available, I scoped the interface around three moments: formulating a question, evaluating an answer, and inspecting its interpretation. A full analytics platform would have diluted the experiment.
Example questions helped make the capability visible. Tables, charts, and short explanations offered different ways to inspect the result.
The experience
The transparency panel exposed the interpretation and query, while keeping correctness and appropriate reliance open to validation.
A transparency panel exposed how the question was interpreted and the underlying database query. It gave stakeholders a way to inspect the result and helped the AI team investigate outputs.
An exposed query is not a guarantee of correctness, and technical detail is not automatically an explanation that every reader can use. That distinction is part of my continuing interest in AI transparency and appropriate reliance.
Validation & iteration
Early stakeholder feedback supported further exploration, with production adoption and accuracy still unestablished.
Early stakeholder feedback supported further development. The experiment gave the team enough signal to continue exploring the concept.
Broader use, accuracy, and the usefulness of explanations still needed validation. This was a freelance proof of concept, not a deployed clinical system.
Outcomes & impact
Early stakeholder feedback supported further exploration of the prototype, while production adoption, clinical accuracy and patient outcomes remained unestablished.
- A prototype for asking questions, reading answers, and inspecting the underlying query.
- Early stakeholder feedback supported further exploration.
- Production adoption, clinical accuracy, and patient outcomes were not established.
