Symptomatik
An AI lab-results explainer where the medically critical decisions are made by code, and the language model only explains.
My role: Product direction, instrument selection and scoring, and the content and SEO architecture.
The Challenge
People get lab results they cannot read and search for answers that are either generic or alarming. An AI explainer is the obvious product, and also the risky one: a language model that misreads a critical value, or invents a reference range, can do real harm. The platform also had to work in three languages from the start.
The Approach
The explainer splits the work. Code classifies every marker against curated reference bands, flags critical values for escalation, and supplies the only facts the model may use. The language model then explains those facts in plain language, and if anything fails, the page falls back to a safe template. Around it sit validated mental-health self-assessments with proper scoring, health calculators, and long-form guides to common lab tests, all in English, Polish and Spanish.
Results
- Lab Results Explainer live in three languages, with a curated set of 25 markers clinically verified
- Deterministic safety core: band classification in code, critical-value escalation, grounded-only facts, safe-template fallback
- 15 validated mental-health self-assessments in English, Polish and Spanish, with crisis-alert handling
- Five health calculators (BMI, blood pressure, HbA1c, cholesterol ratio, eGFR)
- Long-form guides for 98 lab tests, with references
- A routing layer across several model providers, with fallback and cost tracking
The design choice that matters in Symptomatik is where the model is allowed to decide. In health, the answer is: nowhere critical. Whether a value is low, normal, high or dangerous is settled by code against reviewed reference bands before the model sees anything. The model’s job is to explain a decision that has already been made, in words a worried person can follow.
Built through Digital Savages.