From Reflection to Prototype: Building “Interpretive Doctor”
One of the most difficult questions that arises as we face illness- especially a terminal one- is how we want to spend our final days.
We often think we have clarity about what matters most, but do we truly know?
Do we prioritize the longest life possible, using every treatment available to fight the disease? Or are we more focused on minimizing pain and spending our remaining time with family?
Perhaps the answer lies somewhere in between, and understanding what we truly want is the key to making that choice. The shape of our final time is something we rarely think about until we have to.
For me, this uncertainty became clear when thinking about my grandmother. I asked myself, what would she choose if she had to decide her treatment options and end-of-life care? Would she want everything to be done to keep her alive, or would she prefer to spend her last months at home, surrounded by loved ones, in comfort?
These are difficult questions, and answers don’t come easily. But clarifying our values and priorities before we're in the midst of a medical crisis - before we are forced to make those decisions- could make all the difference. And that’s where Interpretive Doctor came from.
From Idea to Requirements
The core insight was simple: patients and families need clarity and agency.
I translated that into three product goals:
Value Ranking – help patients articulate what matters most when treatment choices are difficult.
Question Builder – generate and save tailored questions for a doctor’s visit.
Terminology Simplifier – turn the dense language of prescriptions, discharge notes, and lab reports into plain words.
Those features became the backbone of Interpretive Doctor.
Designing with ChatGPT
Instead of jumping straight to code, I spent time here—inside ChatGPT—drafting prompts and iterating on the concept.
Through a long conversation I refined:
Data models for users, value profiles, treatment options, question sets, and uploaded medical notes.
Page flow: onboarding with drag-and-drop value cards, fit-score visualization, a question list with star/reorder functions, and a public share link for clinicians.
Tech stack: React + MUI for the interface, Supabase for authentication and storage, and OpenAI for language generation.
Each round of prompting produced clearer architecture and even sample code, which I later moved into the build tool.
Rapid Prototyping in Lovable
To turn the plan into something real I used Lovable, a platform that turns natural-language instructions into a full-stack app.
Master Prompt – I pasted the detailed prompt from ChatGPT, describing every feature and the database schema.
Automatic Scaffolding – Lovable generated a React + Supabase codebase in minutes: routes, tables, and starter pages for
/onboarding,/questions,/fit, and/share/:token.Conversational Iteration – Whenever I needed tweaks- larger text for older users, improved drag-and-drop, extra database fields—I just typed new instructions. Lovable regenerated the code and live preview instantly.
Deployment – After connecting my Supabase instance and OpenAI key, a single click deployed the first MVP to the web.
This process let me show a functioning demo instead of static mock-ups within a day.
