Stealth Digital Health Platform, V1

I served as Founding Engineer for a stealth digital-health startup, delivering V1 of its mobile and web platform.

The product gives patients a structured, patient-authored health record covering all aspects of their medical history. It then connects that record to clinically grounded health information, designed to help people understand and prepare for clinical conversations without providing diagnoses or dosing advice.

The platform needed to move quickly while protecting sensitive health data and maintaining clinical traceability. Key constraints included:

Health content needed grounding in authoritative sources such as NICE and NHS material.

The AI assistant had to explain information without crossing into personalised medical advice.

Patient data required ownership controls, GDPR erasure support, and separation from marketing PII.

A small engineering team needed an architecture that could grow without turning into a monolith.

As part of a small development team, I built

Public site, patient app, and doctor-facing experience in a TypeScript monorepo using Next.js, React

Delivered the V1 mobile and web platform, alongside an internal clinician and admin portal.

Established shared packages for design tokens, UI, GraphQL types, authentication, email, and hooks.

Built a GraphQL backend on AWS AppSync with Cognito JWT authentication.

Designed a single-table DynamoDB core with GSIs, TTLs, and point-in-time recovery, plus isolated storage where needed for marketing PII.

Implemented a transactional outbox pattern feeding DynamoDB Streams, EventBridge, and asynchronous workers to avoid dual-write failures.

Structured AWS CDK infrastructure into composition-root stacks and domain modules, allowing new features to be added without expanding a single central stack.

Built secure backend-for-frontend routes for public forms, with same-origin checks, Zod validation, honeypot protection, minimum fill-time checks, and rate limits.

Built grounded health-article generation with Step Functions and Amazon Bedrock, using authoritative source material, structured outputs, validation, citations, and a separate style-review step.

Built a conversational article assistant delivered asynchronously through GraphQL subscriptions.

Added guardrails against personal diagnosis and dosing advice, PII anonymisation, versioned prompts, feature flags, atomic rate limits, and fallback responses.

Implemented knowledge retrieval from source content stored in S3 and indexed in OpenSearch Serverless through a Bedrock Knowledge Base.

Added server-side registration controls, conversation ownership checks, a dry-run-first GDPR erasure workflow with audit history, and private S3 buckets accessed through presigned URLs.

Built an AI-generated health-summary snapshot that patients can share through revocable, expiring links.

Created a product analytics pipeline from application events through Firehose, S3, Glue, Athena, and Grafana.

V1 shipped as a production-ready digital-health platform with web and mobile experiences, a clinician/admin portal, event-driven AWS backend, safety-focused AI features, and privacy controls designed for sensitive patient data.