Healthcare AI: helping professionals find tests and therapies
Project overview
I designed and built an AI knowledge assistant for a healthcare company, with input from scientists, quality control and medical professionals. Employees and external experts, doctors and therapists use it to explore the company’s available tests and suggested therapies in the context of real patient information.
- My role
- Complete design and implementation
- Domain input
- Scientists, quality control and medical professionals
- Users
- Employees and external doctors, therapists and experts
- Reported benefit
- Time savings, greater independence and more complete suggestions
Challenge
The company’s testing and therapy information needed to be useful both to internal employees and to external professionals considering individual patient contexts. Finding and bringing together relevant information could require help from company staff, limiting how much professionals could resolve independently.
Solution
The assistant makes the company’s knowledge searchable through a conversational interface. Professionals can explore relevant testing options and the therapies the company provides, then take questions the assistant cannot answer to employees. Its role is to support professional enquiry and judgement; clinical decisions remain with the medical professional.
Technical Approach
I owned the design and implementation, working with scientists, quality control and medical professionals on the domain requirements. The RAG pipeline chunks and embeds the company’s reference material, retrieves relevant content from PostgreSQL vector search and supplies it to the model with source references. Retrieval and citations support verification; they do not guarantee that every answer is correct or complete.
Data protection and AI governance
Patient identifiers are removed before information reaches the AI model. This is a practical data-minimisation measure supporting the data-protection principles of the GDPR and Switzerland’s Federal Act on Data Protection (FADP).
Source references help professionals check the assistant’s answers, and clinical decisions remain with the medical professional. These measures support transparency and human oversight in the AI workflow, relevant to the EU AI Act. The applicable obligations depend on the system’s intended purpose and risk classification.
Reported benefits
Users reported considerable time savings for both company employees and medical professionals. External professionals can find more answers independently and contact employees when questions remain unresolved. Feedback also described more complete analyses and suggestions than before. These outcomes are based on qualitative user feedback.
Why This Matters
This work connects technical implementation with the people who hold the domain knowledge and the people who use it. My responsibility covered the complete solution, while scientific, quality and medical input shaped how it served a real workflow. It is an example of the AI leadership I offer: understand the work, build with the specialists and support use in practice.
Discuss a similar challenge