Bring the record together
Combine into a single view information scattered across record data, clinical reports and related documents.
Case study · Leading hospital
Medical AI Agent: Bringing the Patient Journey Together in One Summary
A medical AI agent that retrieves information from the patient record and presents it to anesthesia teams as a chronological summary. We designed and built the RAG-based assistant and the interface that makes it available to medical teams. It supports reading the record; clinical judgment stays with the clinicians.
Client
Leading hospital
What we built
RAG assistant and review interface
Use
Support for reading and summarizing the record
Technologies
RAG, large language models (LLMs), business application
The client
Rich records, scattered across many sources.
A leading hospital wanted to make things easier for its anesthesia teams when preparing a consultation.
The information they need exists, but it is spread across record data, clinical reports and documents from the patient journey. Bringing it together takes time, especially with large records.
The challenge
Make the record faster to read, without making it say more.
Put the patient journey in order
Present information chronologically, for a more direct read before the pre-anesthesia consultation.
Stay faithful to the sources
Every item in the summary must come from the record, not from the model’s general knowledge.
Reduce manual searching
Cut the time spent looking for information in large records.
- 01Patient record
Medical record data and related documents.
- 02Retrieval & structuring
Retrieval of relevant passages and organization of the information.
- 03Chronological summary
The patient journey presented in a dedicated interface.
What we built
Medical AI agent: from document sources to a business interface.
We designed and built the entire service, from retrieval in the record data to the screen the teams use. The assistant is not an isolated demo: it is connected to an application designed for the way medical teams work.
Record retrieval
A RAG-based assistant that finds the relevant passages in the record data and documents.
Structuring the summary
Chronological organization of the patient journey.
Delivery to the teams
A review interface that makes the summary available to anesthesia teams.
Health data and evaluation
The principles we apply to a medical assistant.
An assistant that reads patient records processes health data. On this kind of project, we work to the following principles, defined with the hospital and its teams (IT, data protection officer, clinical leads).
Data minimization
The assistant only accesses the information needed for the expected summary.
Access rights
Only authorized clinicians can view the summary, according to the hospital’s access rules.
Hosting and traceability
The solution runs on a secure server certified for health data hosting (HDS, the French certification for health data hosts). Logging and retention periods are defined with the hospital, in line with the regulations that apply to health data.
Evaluating the summary
Check the summary’s faithfulness to the sources and look for omissions on reference records reviewed by clinicians, before every change.
The human role
The assistant helps read and summarize. It makes no diagnosis and no recommendation; the decision stays with the clinicians.
This is the approach described in our LLMOps and AI evaluation expertise and in our article on AI precautions, risks and compliance.
Team and organization
A dedicated team, from data to interface.
The project was delivered as part of an “AI agents” engagement by a dedicated team, the way we deliver: engineers who design, build and evolve the service as the medical teams’ needs emerge.
The same team covers document retrieval, model integration and application design. It therefore understands both how the AI behaves and the screen clinicians use to consult it.
Technologies used
RAG, language models and a business application.
RAG (retrieval-augmented generation) grounds the summary in the patient’s documents rather than in the model’s general knowledge. Relevant passages are first retrieved from the record, then a large language model (LLM) organizes them into a chronological summary.
The architecture connects three building blocks: access to record data, document retrieval and summary generation, exposed through a business interface. It is the same kind of work we do in our RAG development and AI agent projects.
AI agents & chatbotsAgents connected to your documents and tools, built all the way to production.↗RAG developmentAnswers grounded in your documents, with verifiable sources.↗Integrate AIBusiness assistants, document search and automation: AI solutions connected to your data and tools.↗Custom softwareThe application that makes AI output usable in teams’ day-to-day work.↗
Outcome
A summary of the patient journey, in a dedicated interface.
- Anesthesia teams review the patient journey in a single interface, in chronological order.
- The summary is grounded in the record’s documents, not in the model’s general knowledge.
- Preparing a consultation involves less manual searching across sources.
- Clinical judgment remains entirely with the healthcare professionals.
This project shows what we do with AI: connect existing data to a concrete use, then build the software that makes that information usable. We apply the same approach in-house with Bobby, our AI assistant connected to our ERP.
Frequently asked questions
Medical AI agent: your questions.
What does the medical AI agent we built do?
It retrieves information from a patient’s medical record and presents it as a chronological summary, in an interface built for anesthesia teams. It makes the record easier to read and prepare.
Does the agent replace medical decisions?
No. It supports reading and summarizing the record. The assistant makes no diagnosis and no recommendation; clinical judgment and decisions remain with the healthcare professionals.
Why use RAG for a medical AI agent?
RAG grounds the summary in the patient’s documents. The assistant first retrieves the relevant passages from the record, then the language model organizes them. The summary stays tied to the record’s sources. See our RAG development expertise.
How do you handle health data on this kind of project?
Through principles defined with the hospital: data minimization, access limited to authorized clinicians, hosting on a secure HDS-certified server, and logging and retention in line with the regulations that apply to health data.
How do you evaluate the quality of a medical summary?
On reference records reviewed by clinicians. We check faithfulness to the sources, omissions and readability, and we repeat this evaluation before every change. See our LLMOps and AI evaluation expertise.
Can you build an AI assistant on our own documents?
Yes. That is what our enterprise AI solutions and AI agents are for: connecting a model to your document sources, building the application around it and supporting the production release.
A document collection that could be more useful.
Tell us about your users, your software and what you want to build or improve.