Services
Artificial intelligence
AI features that do real work inside a product — image, document and text analysis, agents and assistants.
AI inside a product is not a demo — it is a feature that has to work every time, in several languages, at a controlled cost. In Healthness a user photographs a meal and gets a nutritional breakdown: the model call goes through the server, keys stay server-side, the response is structured and validated, and the quota is metered per user. The same approach applies to documents, quotes, customer support and search over your own content.
Technology
- Claude API
- OpenAI API
- Vercel AI SDK
- AI Gateway
- Vision
- RAG
- pgvector
- Cloud Functions
What you get
- Photo and document analysis (invoices, labels, forms)
- An assistant and chat over your own data and documentation (RAG)
- Automatic classification, summarisation and data extraction
- AI agents that carry out multi-step tasks and call your tools
- Structured, validated responses — no invented fields
- Quotas, limits and per-user cost metering
- Keys and secrets kept strictly server-side
- Quality evaluation before anything goes live
Proof in the work
Healthness
iOS and Android app · own product
An app for fasting, nutrition and daily rhythm. Two native platforms over one backend, three languages, subscriptions and AI meal photo analysis — all built, shipped and maintained solo.
- Swift
- SwiftUI
- Kotlin
- Jetpack Compose
WATdesign
Catalogue and configurator · web
A made-to-measure wall mural catalogue with 4,620 designs, a dimension configurator and per-m² pricing — with no database and not one image file in the repository.
- Next.js 16
- React 19
- TypeScript
- Tailwind CSS 4
What does an AI feature cost per month?
It depends on the number of calls, and that number is something you plan. Per-user quotas and spending limits are built in from day one, so the monthly bill is predictable instead of a surprise. Before launch you get a cost estimate per user and per call.
Does our data stay private?
Only what the task requires is sent, the call goes from the server rather than the browser, and providers and settings are chosen with zero data retention for training. Exactly what is sent is written down in the documentation you receive.
Contact
Tell me what you are building — I reply today.
I normally reply the same working day. The first call takes thirty minutes, costs nothing and commits you to nothing.