ChiraYou is a German healthtech startup focused on long-term, preventive health tracking. The founder reached out on a referral. He had a clinical team, a lab partnership, and a working backend prototype, but the user-facing product did not exist yet — no mobile app, no doctor portal, no checkout flow, no AI pipeline.

The founder had a clear product vision but no detailed spec. Most of the product and engineering decisions on the client-facing side were up to us. Here is what the product needed to do by the end.

• One app for the full preventive health journey (ordering a test, generating the quarterly plan, everything in between) on iOS and Android from one codebase.

• Useful output from Gemini across 20+ biomarkers in four categories (fitness, reserve, reset, flow), with medical history, family risks, wearable data, and age- and sex-adjusted reference ranges feeding into it.

• A workspace for doctors to review patient data, comment on it, and schedule consultations.

• Localization at the architecture level so the app, AI reports, emails, and push notifications all work in the user's language.

The project came to us through a referral. The founder needed a team that could take ownership of the entire product layer and move forward without waiting for detailed specs.

1. ONE TEAM ACROSS THE WHOLE PRODUCT

Mobile app, backend, and two web portals, all covered without handing work between specialists from different stacks. For a startup at this stage, it meant faster decisions.

2. EXPERIENCE WITH SENSITIVE DATA

We had already built products handling medical records and personal information. The founder did not need to explain why certain things in healthtech cannot be done the way they are done in regular SaaS.

3. WORKING WITHOUT A RIGID SPEC

The product was evolving as we built it. We took part in product decisions and pushed back with alternatives whenever the original plan stopped holding up.

4. HANDS-ON EXPERIENCE WITH AI IN PRODUCTION

We had shipped LLM-based features before, so we came in with a sense of what to do and what to avoid.

ChiraYou is a product in a regulated industry where AI output shapes decisions about a person's health. The usual development playbook does not apply here. We did not build the model ourselves, we integrated Google Gemini. But integrating an LLM into healthtech is very different from a typical product, and every requirement of the industry turned into its own engineering problem.

1. BRINGING DIFFERENT TYPES OF MEDICAL DATA TOGETHER

The data going into the AI came from many places. Lab tests from Tasso+, blood pressure monitors and grip dynamometers, wearables connected through Apple HealthKit and Google Health Connect, cognitive tests inside the app, manual entries. Every source had its own units, its own update frequency, its own reference ranges that depend on the patient's age and sex. Feeding raw data of that shape into a model gives you a mess on the other end.

Solution

We added a normalization layer that every input passes through before the AI sees anything. Lab values get converted into one consistent set of units. Reference ranges are pulled in dynamically based on the user's profile. By the time the data reaches Gemini, it is one clean structured object.

2. THE HEALTHSPAN SCORE ACROSS 20+ MARKERS

The product's main metric is the number of additional healthy years a user can expect. On the home screen this shows up as a single number. Behind it sit 20+ markers across four categories, plus medical history, hereditary risks, and current medications. A simple average will not do. Strong markers in one category can hide a serious risk in another.

Solution

We built a pipeline where Gemini receives the markers along with structured patient context. The prompt is set up so the model explains the result to the user and produces a list of 10 priority actions for that specific cycle.

3. AN ADMIN PORTAL BUILT FOR A REGULATED INDUSTRY

A basic user management panel is not enough in healthtech. The clinic needs doctor verification, consultation management, audit logs, billing, contracts, transcripts. Without these, the team cannot operate and the product cannot meet medical data handling requirements.

Solution

We built a separate portal on React + Vite with a full audit trail. Every action in the system gets logged with a timestamp, an event type, and the user who performed it. We added separate roles for doctors and admins, filters across consultations, and data exports. The whole thing runs on Google Cloud Platform. That covered EU data residency and most of the access controls we needed at the infrastructure level.

4. LOCALIZING MEDICAL CONTENT

Translating the UI is the easy part. The harder problem is getting Gemini to write recommendations directly in the user's language. Forcing the model to write in English and then translating produces unnatural phrasing that reads wrong in a medical context.

Solution

The user's language became part of their profile and gets passed through every layer of the system. The prompt to Gemini includes an instruction to write the report in that language. PDF templates, push notifications, and email campaigns pull translations from the same variable. Adding a new language is a small change, not a product rewrite.

5. AI ASSISTANTS IN DEVELOPMENT

A team of two engineers covered an 11-month scope that would normally take 4 to 5 people. Mobile, backend, two web portals, AI pipeline, lab integration. Without AI tooling, this would not have fit into the timeline.

Solution

We leaned on LLMs heavily during the early product thinking. When we had a rough idea for a feature or a flow, we would talk it through with the model first, ask it to find weak points, and use it to put together a quick PoC. That meant we could test ten ideas in the time it would normally take to manually validate one. The team only spent real engineering time on concepts that already held up. Later in the cycle, we used AI for first drafts of unit tests, Maestro E2E scenarios, and pull request reviews. The model did not work with code that handles medical data, did not write Stripe logic, and did not make architectural decisions.

Frontend: React Native, Expo, React, Vite, Tailwind CSS, shadcn/ui, TypeScript

Backend: Node.js, Express, tRPC, Puppeteer (PDF and invoice generation)

Database and infrastructure: MongoDB, Google Cloud Platform, Google Cloud Storage

Third-party integrations: Google Gemini, Stripe, Mailgun, Swiss Analysis, Apple HealthKit, Google Health Connect

In 11 months, ChiraYou went live with a full product stack ready for EU markets.

• Mobile apps, doctor and admin portals, and a standalone checkout running at chirayou.com.

• Doctors have run 150+ consultations through the platform since launch.

• Every AI report reaches the user only after a doctor reviews it.

• The product runs in multiple languages across UI, AI output, emails, and push notifications.

• Two engineers delivered the full stack in 11 months.