Generative AI for cardiologists & residents

Not a marketing story, this is how LLMs work, what you can do with them in clinical cardiology, and why you are the safety layer. Six modules with the same structure as ChatGPT for beginners: short explanation, 2–3 multiple choice questions, and a live exercise you do directly in ChatGPT or Claude.

6 modules · ~7–9 hours · clinically applicable from day 1
LLM fundamentals, tokens, embeddings, transformer, attention, context window
The players, ChatGPT, Claude, Gemini (plus DeepSeek) and which to pick when
Cardio prompt architecture, DDx for chest pain, discharge letter, patient leaflet, echo, teaching
Risks & GDPR, traffic light model, hallucinations, bias, the four rules of thumb

Start the course View modules

For cardiologists and residents with basic familiarity with ChatGPT or Claude. Works on pc, tablet and phone.

Cardiologist working with AI assistant in practice

What you'll be able to do

Explain what an LLM is in one sentence

Tokens, embeddings, attention and context window, not as jargon but as explanation for concrete prompt rules of thumb.

Make good clinical prompts

Role + context + task + format + uncertainty. With templates for DDx, discharge letter, patient leaflet, guideline summary, teaching.

Compare the players

ChatGPT (GPT-5.6), Claude (Fable 5.1 / Opus 5) and Gemini (3.1 Pro / 3.8 Flash), strengths, pitfalls and the right choice per cardio task.

Recognise and prevent hallucinations

Why an invented table number isn't a bug, and how to keep it out by “providing source instead of asking source”.

GDPR & privacy traffic light

Green, amber, red. In 5 seconds know if something belongs in a public chat, and when you need a hospital-approved environment.

Be the safety layer

Recognise automation bias, four rules of thumb, and local governance (DPIA, ESC AI Hub, EACVI). Final accountability stays with you.

The six modules

Module 1 · foundation ~60 min

What is an LLM?

AI → ML → DL → LLM. The core trick in one sentence. Tokens and embeddings. Reasoning with meaning (why RAG works). Scale and emergent behaviour, why today's models can do things that failed last year.

Go to overview

Module 2 · under the hood ~75 min

How is an LLM made?

Pre-training, supervised fine-tuning, RLHF, DPO. Domain-FT (Med-PaLM, MedGemma) versus RAG. The Transformer and attention. Context window, why you put your question at the bottom and what “lost in the middle” means for long ESC guidelines.

Go to overview

Module 3 · tools ~60 min

The players, opportunities for the cardiologist

ChatGPT, Claude (and Anthropic's ethical positioning), Gemini and DeepSeek. Which for which task: letters, full guidelines, code, image, voice. Cardio choice matrix. The exponential pace and what it means for you.

Go to overview

Module 4 · clinical practice ~90 min

Cardio prompt architecture

The five building blocks (role/context/task/format/uncertainty). Provide source instead of asking source. Five live exercises: DDx for chest pain, discharge letter from dictation, patient leaflet B1, echo summary for MDT, teaching with case + learning questions.

Go to overview

Module 5 · safety layer ~60 min

Risks, and how to frame them

Hallucinations as a feature, GDPR traffic light (green/amber/red), bias in the model and in yourself (automation bias), the four rules of thumb and local governance (DPIA, ESC AI Hub, EACVI). Final exercise: hallucination hunt.

Go to overview

Module 6 · case module ~60–90 min

The big case, cardiogenic shock after anterior wall infarction

One realistic case adapted for teaching (Impella, VA-ECMO, LVAD screening) on which you apply everything: summarising in three formats, DD sparring, fact-checking, treatment plan vs. guideline, prompt iteration & privacy. Five work assignments, ten multiple-choice questions.

Go to overview

What colleagues say

Initial reflections from cardiologists and residents who took the course. New reviews are added here regularly.

Finally a course that doesn't end at “ChatGPT exists” but actually goes into tokens, attention and what it means for my prompt work. The DDx exercise was very concrete.

Cardiologist, peripheral hospital

Module 3 was eye-opening: I didn't know Claude can do 200K tokens and I only used ChatGPT. Now I deliberately pick a tool per task. Much calmer work.

Cardiology resident

The GDPR explanation and the four rules of thumb are exactly what our department needed. Mistral as the EU route was a surprise, I'm bringing it to our IT working group.

Cardiologist

Frequently asked questions

Who is this course for?

For cardiologists and cardiology residents who already have some experience with ChatGPT or Claude and want to understand the technology better to deploy it more responsibly in clinical practice.

Do I need ChatGPT Plus or Claude Pro?

The course works with the free versions. For long contexts (Claude 1M tokens, Fable 5.1) a paid account is sometimes useful but not required for the exercises.

May I put patient data into ChatGPT?

Module 5 covers this in depth. Short rule up front: no patient data in a public chat. Anonymised or fictional cases are fine. For real data: only in a hospital-approved environment (Microsoft Copilot M365 business, Azure OpenAI, Claude Enterprise, depending on DPIA).

Is this an ECG interpretation course?

No. We discuss how to put ECG and echo findings from text into a prompt and use them for letters, DDx or teaching. ECG interpretation itself stays with you or a specialised ECG AI system (PMcardio, Cardiologs, etc.).

What if I get stuck?

The modules are designed for your own pace. Every “Live exercise” block gives a ready-to-use prompt with links to ChatGPT and Claude. Content feedback on the course itself? Mail us via the portal contact page.