Inspiration
Pain is subjective, data should not be. Patients get a 15-minute appointment every few months and are asked to describe 60+ days of experience on a 1-to-10 scale. Critical patterns get lost. PainPoint captures them while they're fresh and hands them back when you need them most.
What it does
Tap a region on the body, pick a pain type, slide the intensity, note the trigger. Under 15 seconds per log. Over time, the app builds a spatial-temporal profile showing where pain concentrates, how intensity trends, whether it's migrating between regions, and which triggers dominate. One button generates two things for the appointment: a clinical summary a doctor can scan in 20 seconds, and a list of specific questions to bring in, tied directly to your own data.
How we built it
React single-file app, custom SVG bodies for front and back views, localStorage for persistence, no backend. Stats, trend detection, and heatmap rendering all client-side. The AI layer calls Gemini 2.0 Flash through Lovable's portal, stored locally. A single prompt built from the computed stats returns both the doctor summary and the patient questions as JSON. Hardcoded fallbacks kick in if the AI call fails. Design language is intentionally calm, not clinical: beige and cream palette, sage accents, Playfair Display headings, rounded cards.
Challenges we ran into
The body SVG was harder than expected. Early versions looked like a biology textbook, which is the exact clinical energy we were trying to avoid. The prompt also went through heavy revision. First drafts over-diagnosed or gave generic "talk to your doctor" hedges. We rewrote the instructions to force the model to lead with the most diagnostically significant pattern and never recommend treatment. Trend detection on sparse data was also tricky. With 30 entries across 3 weeks you can't run a regression, so we compared rolling averages from the first and last few days of the window.
Accomplishments that we're proud of
The 15-second capture actually holds up. The doctor summary reads clinically, like something a medical student would hand a physician, not an AI dump. And the advocacy layer, generating patient-specific questions grounded in the actual log, is something we haven't seen in any other pain tracking app.
What we learned
The hard part of a health app isn't the tech, it's design restraint. We could have added mood tracking, sleep correlation, medication logs, flare prediction. Every feature we cut made the app better for the person it's actually for. We also learned AI is most powerful when scoped narrowly: a general "tell me about my health" prompt produces mush, but "lead with the most diagnostic pattern, flag migration, never recommend treatment" produces something a doctor can use.
What's next for PainPoint
We are planning to execute a test phase where we get a solid base of test users. From there we can implement "ease of access" features and improve our app. Long term, we plan on pushing this out to clinics and having doctors recommend the app to users through the App Store(0.99). Long term we see a real B2B data opportunity. Clinical trials and pharma measure pain through infrequent surveys, and wearables can't sense subjective pain at all. High-frequency, patient-logged pain data is a category of evidence that doesn't really exist in structured form. Longer term: native iOS with Apple Health integration, medication response tracking, and condition-specific prompt tuning for endometriosis, fibromyalgia, migraine, and post-surgical recovery.
Log in or sign up for Devpost to join the conversation.