DentaSmart answers a simple, stressful question: what is actually going on in my mouth? Users photograph their teeth or upload X-rays, and within minutes the platform returns color-coded findings, an oral health score, and two reports: a plain-English one for the patient and a clinical version they can hand to their dentist. The product has analyzed more than 50,000 scans and holds a 5.0 rating on the App Store.
I lead backend development on this product, owning the architecture from the API gateway down to the model serving layer.
The problem
Dental care is a black box for most people. Jargon makes patients feel powerless, costs appear after they have already said yes, and by the time something hurts it is usually expensive. The founding team wanted an app that gives people a clear starting point before they sit in the chair, without pretending to replace a licensed dentist.
Technically, that meant a system that could ingest photos and X-rays from mobile devices, run them through an in-house trained vision model, translate raw model output into language a patient can act on, and do all of it fast enough to feel instant.
What I built
The platform runs as a set of services with clear boundaries. Django powers the core domain: accounts, scan history, reports and the employer program. FastAPI handles the inference pipeline, where speed matters most. NestJS services sit closer to the client apps, aggregating data for the web dashboard and the iOS and Android apps.
The image pipeline is the heart of the product. Uploads land in object storage, a Celery worker picks them up through RabbitMQ, and the serving layer I built runs them through our custom-trained model. Findings come back as structured data with confidence levels, which the report generator turns into the two-audience output: friendly for patients, precise for clinicians.
On top of that sits an LLM assistant that uses retrieval-augmented generation over a vector database of vetted dental content. When a user asks about a finding, the assistant answers grounded in our reviewed material rather than improvising, which was a hard requirement from the clinical advisory board.
The details that mattered
Money and identity had to be boring and reliable. I implemented OAuth2 flows across all clients and built the Stripe integration for subscriptions, including the employer plans where a company covers scans for its staff. Everything deploys through Docker-based CI/CD to AWS, with the inference service scaled separately from the web tier because their load profiles are completely different.
Where it stands
DentaSmart is live on the App Store and Google Play with a free tier, processing scans around the clock. The clinical report feature has become a genuine differentiator: users walk into appointments already understanding their situation, and dentists receive a structured summary instead of a worried guess.