Preventive Medicine Application

Preventr turns your health data into an actionable health score with a personalized prevention plan—showing exactly what matters most, and what to do next.

Comment

Inspiration

As a Preventive Medicine physician, I understand that preventing diseases from happening or getting worse should be the main focus of the healthcare system worldwide. Yet, it is a type of Medicine that does not generate as much revenue and tend to be set aside. My goal is to make Preventive Medicine much more accessible to everyone with guidance across lab results, lifestyle habits, screening schedules, vaccine history.

The Preventive score is personalized, and everybody will be able to get the perfect score, according to their genetics, physiology, and resources.

It will be clear to them:

  1. Why is the score what it is?
  2. What should they do next?

What it does

Preventr turns fragmented health information into a clear, personalized prevention plan.

Members complete a five-section assessment covering cardiometabolic health, movement, nutrition, sleep, substance exposure, cancer screening, and immunizations. Preventr then produces:

  • One Core Action Score and seven health-area scores
  • A separate completeness measure, so missing information is never treated as poor health
  • A point-by-point explanation of what shaped the score
  • The current value, target, available points, rationale, evidence, and next action for every deduction
  • A ranked plan showing the three most valuable next steps
  • Searchable health data and unit-safe trends
  • Personalized average-risk screening guidance
  • An AI Health Companion designed to explain approved health context without calculating the score, diagnosing, or prescribing

The public demo works without an account or personal health information.

How I built it

Preventr is a responsive web application built with React 19, TypeScript, and a Next-compatible Vinext app router. It is deployed on Cloudflare through OpenAI Sites. All things that I let Codex with GPT 5.6 Sol take care of, since I am not familiar with any of those terms.

The scoring engine is written as TypeScript. Its rules are versioned, testable, and independent from AI. Every score deduction can be traced from the member's current value to the applicable target, available points, explanation, next action, and evidence source.

Supabase provides the authentication and persistence foundation. The database design uses PostgreSQL Row Level Security to protect assessments, health observations, entitlements, and score-version history.

GPT-5.6 in Codex was my primary development collaborator. It helped turn the product brief into a working architecture, implement the assessment and dashboard, test clinical and scoring edge cases, establish the Supabase and AI trust boundaries, and build the automated verification suite.

Free and Plus tiers use the same score, missing information is not silently imputed, urgent safety information is never paywalled, and future capabilities are never presented as live.

Challenges we ran into

The hardest problem was not designing the dashboard. It was preserving trust across information with different units, evidence strength, eligibility rules, and missing-data behavior. I would say that the current scoring system is still in the beginning and it requires further validation.

A few examples:

  • Select one score-driving lipid measurement when ApoB, non-HDL cholesterol, and LDL-C are all available and can say different things about your health.
  • Keep Lp(a) unit-specific and outside the score
  • Separate BMI scoring from body-fat interpretation
  • Exclude genuinely non-applicable screenings without awarding artificial points
  • Evaluate routine Td/Tdap history without inventing vaccination timing
  • Distinguish missing information from an unfavorable result
  • Keep guest recovery convenient without presenting browser storage as durable health-record storage
  • The AI companion requires training and API costs. It is added just as an example for now.

These constraints required scoring, safety, data architecture, and interface language to be designed together.

Accomplishments that I am proud of

  • A complete consumer experience rather than a scoring proof of concept
  • Transparent, deterministic scoring across seven areas of health
  • A practical action plan generated from every score deduction
  • Clear separation between score and data completeness
  • Honest safety, applicability, demo, account, and error states
  • Direct section editing without repeating the full assessment
  • Unit-safe health trends that do not invent continuity
  • A server-controlled GPT-5.6 context boundary with explicit medical-safety limits
  • A no-login path that lets judges explore the complete product using demo data

What I learned

A health score becomes more useful when every lost point can be converted into a transparent receipt and a realistic next action. Missing information must also remain visibly different from a low result.

We learned that AI is strongest in this product as an interpretation and workflow layer. The canonical score is safer, more reproducible, and easier to verify when it remains deterministic.

What's next

Before clinical production use, Preventr will require formal clinician validation, complete privacy and consent materials, production safety evaluation, rate limiting, and live-infrastructure testing.

The roadmap includes native Apple Health and Android Health Connect integrations, lab ingestion, clinician-ready visit briefs, adaptive weekly plans, deeper longitudinal insights, and common EHR synchronization.

Built With

Share this project:

Updates