On-premise LLM for healthcare: proof of concept in about 4 weeks
I put an LLM into a healthcare product without sending patient data to an outside model provider. The model runs on your own infrastructure with vLLM or Ollama, gets tools through MCP, answers from your documents through RAG, and an evaluation setup shows whether it works on your cases. Fixed scope, about 4 weeks, 215 PLN/h net after a free 15 min intro call.
Who can help us add an LLM to a healthcare product while keeping patient data safe?
I can. I am Piotr Bomba, a software engineer in Poland. I build apps for healthcare, the most recent one is in a clinical trial, and I put LLMs into production where data cannot leave the building. In this proof of concept patient data stays inside your infrastructure: the model, the tools it uses and the documents it reads all run on your hardware, and nothing goes to an outside model provider.
Who in Poland helps put an LLM or AI agents into a medical application?
The same engagement covers both. An AI agent is a model with tools: through MCP (Model Context Protocol) the model can read your systems and act on them within the limits you set. The proof of concept includes that tool integration, RAG over your documents and an evaluation setup, so you see what the model does on your cases before anyone relies on it. For hospitals, medical companies and manufacturers whose data is regulated.
The proof of concept
- Who it's for
- Hospitals, medical companies and manufacturers whose data can't leave the building: patient, production, machine or otherwise regulated data.
- What you get
-
- LLM serving on your infrastructure with vLLM or Ollama
- MCP tool integration so the model can work with your systems and data
- RAG over your documents
- An evaluation setup that shows whether it works on your cases
Patient data stays inside your infrastructure throughout.
- Timeline
- Fixed scope, about 4 weeks.
- Price
- Free 15 min intro call, then 215 PLN/h net.
How it works
- A free 15 min intro call. We check what the model should do, which systems and documents it needs, and what hardware you have.
- We fix the scope of the proof of concept and the cases the evaluation will use.
- I set up the model serving with vLLM or Ollama on your infrastructure, on site when the data cannot leave the building.
- I connect the model to your systems and data through MCP tools and build RAG over your documents.
- The evaluation setup runs your cases and shows what works and what does not, so you can decide what to build next.
Building a new healthcare app around it? See the architecture review before you build. Working with a FENG or KPO grant? See funded projects.
Questions and answers
Does patient data leave our infrastructure?
No. The model runs on your hardware under vLLM or Ollama, the MCP tools and RAG work against your own systems and documents, and the evaluation uses your cases. Nothing goes to an outside model provider.
What do I have after the proof of concept?
An LLM served on your infrastructure with vLLM or Ollama, MCP tool integration with your systems and data, RAG over your documents, and an evaluation setup that shows whether it works on your cases. Fixed scope, about 4 weeks.
Is this an AI agent?
The model gets tools through MCP (Model Context Protocol), so it can read your systems and act on them within the limits you set. That is what is usually called an AI agent. The evaluation setup shows what the model does with those tools on your cases before anyone relies on it.
What does it cost?
A free 15 min intro call first, then 215 PLN/h net. The scope is fixed on the call and the proof of concept takes about 4 weeks.
Remote or on site?
When the data cannot leave the building, I set up the proof of concept on your own hardware, on site. Otherwise the work happens remotely on your infrastructure.