Portfolio Project · Active Development

WellMatch

A health aware product recommendation engine that bridges ecommerce and clinical context. The right wellness products, for the right person, safely.

57%
of consumers want health aligned product tools
6 in 10
US adults have at least one chronic disease
$200B
wellness beauty market, growing 4% YoY
The Problem

Wellness shoppers purchase products that
conflict with their health, not from carelessness

Product pages do not surface health compatibility meaningfully. A customer managing a chronic condition, taking prescribed medications, or following clinical dietary guidance has no reliable way to filter a product catalogue against their health profile before purchasing. This creates risk (adverse reactions, wasted spend, eroded trust) and missed opportunity (higher confidence purchases, increased basket size, better retention).

Who It Serves

Four audiences. One engine.

Primary · Consumer
The Health Conscious Shopper
Manages their health actively. Researches carefully. Wants confidence, not just choice.
57%
would use a health aligned product tool (NIQ 2025)
Primary · Consumer
The Care Adjacent Buyer
Purchasing on behalf of someone with health conditions. Stakes are higher. Expertise is lower.
24.4%
of adults 45 to 64 are caregivers making product decisions
Secondary · B2B
The Wellness Brand Operator
DTC wellness brand evaluating WellMatch as a conversion and return reduction layer.
$200B
wellness beauty market, growing 4% year over year
Secondary · B2B
The Healthtech Evaluator
Health platform assessing WellMatch as infrastructure for a health aware commerce module.
74%
prefer tech products with health and wellness features
How It Works

Intake, match, explain.

01
Health Intake
User answers 5 questions or fewer covering conditions, dietary restrictions, and ingredient sensitivities. Skippable. Under 90 seconds.
02
Compatibility Matching
A rule-based matching engine scores each product's actual ingredients against the user's health profile and returns a ranked list with confidence tiers: safe, use with caution, consult a professional.
03
Plain Language Explanation
Every recommendation comes with a plain language reason. Not clinical jargon. Not just a badge. A sentence that tells the user why this product works for them.
Technical Stack

Built on tools I have already shipped with.

React / Vite / Tailwind Deterministic rule-based matcher No external API calls Shopify (v1.5) Vercel Pendo Ingredient-level JSON (v1) Session based, no auth (v1)

Every product is defined by its actual ingredient list, and every ingredient carries structured health metadata: conditions to avoid, conditions to flag, conditions it helps, allergen tags, vegan status. The engine scores each ingredient against the user's profile and rolls the result up to a product-level tier with plain language rationale.

Why rules over an LLM: a deliberate choice, not a shortcut. When the output is a health compatibility signal, every recommendation needs to trace back to an explicit, reviewable rule, not a language model's best guess. A rules engine can't hallucinate a contraindication that isn't in the data. It also runs free, at any volume, with no per-request latency. The tradeoff is real: rules don't scale to a thousand-SKU catalogue as gracefully as an LLM could, but for a safety-adjacent recommendation surface, traceability won over flexibility.

Read the full PRD
Problem statement, goals, user stories, requirements, success metrics, open questions, and timeline.
Open PRD →