Case studies
mojApteczka — an end-to-end product built on GenAI.
Our strongest proof of competence: a complete digital product we designed, built and operate — from the backend to mobile apps.
Context
From zero to a working product.
mojApteczka is a knowledge-driven digital product in the healthcare domain — exactly the type of system companies need most often: trustworthy answers from controlled sources, served safely and at scale. We built it entirely on Generative AI, in a production architecture, not in a data scientist's notebook.
In brief
mojApteczka in one paragraph — facts to cite.
mojApteczka is a complete digital product built on Generative AI that Semitora designed, built and operates on its own — from an AWS cloud backend, through ETL pipelines and a RAG knowledge base with source citations, to iOS and Android mobile apps and AI-driven customer support. The same stack — architecture, data, quality evaluations, GDPR and AI Act compliance — is what Semitora transfers into client deployments.
It runs in production, not in a demo. A June 2026 operating snapshot contained 302,516 drug-interaction records, while the rounded inference-layer cost of one AI scan was USD 0.0006. Those are parameters of this system and point in time, not a benchmark for another deployment. We withdrew the earlier extraction-accuracy percentage because the public material did not retain a reproducible sample, scoring definition and per-case result.
Updated: 30 July 2026 · mojApteczka first-party data and source review.
What we built
The full stack — the same one we deploy for clients.
One working production system, layer by layer: a cloud backend on AWS, ETL pipelines, a RAG knowledge base with citations, AI-powered support and mobile apps. Semitora carries the same set of layers into client deployments.
Cloud backend on AWS
Production architecture: scaling, security, monitoring and inference cost control.
ETL & data
Pipelines transforming source medical data into a structured, versioned knowledge base.
Knowledge bases / RAG
Answers generated exclusively from controlled sources, with citations and quality evaluation.
AI-powered support
User support with AI on the first line and a human in the loop where it matters.
Mobile apps
The product in end users' hands — iOS and Android, with a full release cycle.
Challenges
What was genuinely hard — and what we learned.
Hallucinations in a high-stakes domain
In healthcare, an “almost right” answer is dangerous. The fix: strict RAG, source citations and automated answer-quality evaluations.
Inference costs at scale
A naive GenAI architecture can eat your margin. The fix: caching, matching models to tasks, and per-query cost monitoring.
Messy source data
The biggest work happened before the model, not inside it: ETL, cleaning and versioning the knowledge. Every company faces the same — and we know the way through.
Security and compliance
A high-stakes domain — with control and compliance.
Healthcare forgives neither errors nor leaks. We built these same mechanisms into mojApteczka and deploy them for clients.
Data and GDPR
Data in a controlled, versioned knowledge base; architecture and access designed for GDPR.
AI Act compliance
Risk classification, documentation and oversight — the system is designed for AI Act obligations, not retrofitted after the fact.
Hallucination control
Strict RAG, source citations and automated evaluations — when there is no answer, the system says “I don’t know”.
Human oversight
AI on the first line, a human in the loop where the stakes are high.
The same approach for you: RAG, AI Act compliance and AI for healthcare.
Results
RAG
versioned knowledge, source citations and controlled refusal
$0.0006
cost per AI scan in production
302,516
drug-interaction records in the production knowledge base
mojApteczka production data, June 2026.
We run on it ourselves
We built it — and we run on it.
We don't just advise. We run the mojApteczka brand on AI we built ourselves: content, reach and support — multilingual, almost hands-free.
Multilingual blog
The blog you are reading is our system: posts are produced, reviewed and quality-controlled, and published in many languages — for reach and better SEO. (How: generation + quality control + translation.)
See the blog →End-to-end podcast
A podcast from recording to publication: editing, transcription, social promotion, language versions. (How: production + distribution + transcription.)
Listen →Video on YouTube
We create real videos and auto-publish them to YouTube — another multilingual reach channel. (How: video production + auto-publishing.)
Watch →Multilingual chatbot
Answers customers in their own language, from controlled sources — live on the product site. (How: RAG + multilingual.)
Try it on mojapteczka.pl →Support bot with triangulation
A ticket first reaches a bot: it triangulates the problem, validates the submitted data, tags what was already solved in earlier versions — and hands a complete picture to the fixer bot. (How: triangulation + validation + tagging + handoff.)
AI → human bounded AI workflow with escalation; no rate published without a period and denominator
We will deploy the same approach for you.
Architecture, data pipelines, RAG, evaluations, applications — everything we built for our own product, we transfer into your organisation. No experimenting on your budget.
Further case studies — client implementations in finance, manufacturing, healthcare and hospitality — will be published as projects complete.