Tanja Magoc, Ph.D. 
Take QPSi’s free online training Large Language Models In Clinical Settings to learn more about how to integrate LLMs into your practice and earn free CME credit.
Integrating AI into Your Clinical Practice
Artificial intelligence, particularly in the form of Large Language Models (LLMs), is rapidly becoming a practical tool in professional settings. An LLM is a sophisticated AI trained on vast amounts of text data to understand context, generate human-like writing, and answer complex questions.
For a busy clinician, the applications are immediate and impactful. Imagine being able to summarize a dense new research article in minutes before rounds, or brainstorm different ways to explain a complex diagnosis to a patient and their family. From drafting administrative emails to generating initial talking points for a presentation, LLMs can serve as a powerful assistant, helping you manage information and streamline communication. The key is learning how to interact with them effectively to ensure the results are accurate, relevant, and useful for your specific needs.
Resources
The resources below provide practical guidance for using large language models (LLMs) effectively in clinical and professional settings. You’ll learn the fundamentals of the technology, as well as techniques for generating reliable, high-quality results.
Within the University of Florida, NaviGator provides a secure, private environment for clinicians, faculty, staff, and students to leverage the power of these models for a wide range of language-based tasks. These resources include a foundational understanding of NaviGator and demonstrate how you can begin using it to enhance your daily workflow.
Introduction to Large Language Models
This video defines LLMs and their function in modern AI tools. It introduces NaviGator as the University of Florida’s secure platform for interacting with LLMs and provides practical examples of how a clinician might use this technology—from summarizing research articles to brainstorming patient talking points—to improve efficiency and manage complex information.
Transcript: Introduction to Large Language Models
You’ve likely noticed that artificial intelligence is becoming more integrated into our daily lives.
At the heart of this evolution are Large Language Models, or LLMs. Think of an LLM as an advanced AI‑trained model that understands and generates human‑like text. These models can process vast amounts of information to engage in conversations, answer complex questions, and perform a wide range of language‑based tasks.
You may have already heard of some popular examples such as ChatGPT, Gemini, and Claude. The University of Florida uses NaviGator—a secure, self‑service platform—where students, faculty, and staff can integrate with various LLMs using data in a private environment. The power of these tools extends from our personal lives to our professional responsibilities.
For instance, you can use an LLM to quickly summarize a lengthy research article before your morning rounds, or to brainstorm talking points for a patient‑education session. On a personal level, it can help draft a tricky email or even plan a family vacation. As LLMs become more common, knowing how to interact with them effectively is a crucial skill. Learning the right techniques to guide the model ensures the results you receive are accurate, relevant, and genuinely useful.
This knowledge transforms a novel technology into a powerful tool that enhances your efficiency and simplifies complex tasks. The upcoming series of videos will show you how to harness the power of NaviGator to streamline your workflow.
Note: NaviGator is only available to University of Florida and UF Health employees and students. However, the guidelines presented are applicable to a range of other publicly available AI products.
Basics of Prompt Engineering: The RTF Formula
Learn the essential skill of prompt engineering to ensure your interactions with NaviGator are efficient and accurate. This video breaks down the RTF (Role, Task, Format) framework, a simple method for structuring your requests. By clearly assigning a role, defining a specific task, and specifying the desired output format, you can transform NaviGator from a general tool into a precise clinical assistant. The video also touches on advanced techniques like “chain-of-reasoning” to improve the reliability of complex queries.
Transcript: Basics of Prompt Engineering: The RTF Formula
You’ve likely noticed that artificial intelligence is becoming more integrated into our daily lives. At the heart of this evolution are Large Language Models, or LLMs. But what are they exactly?
Think of an LLM as an advanced AI‑trained model that understands and generates human‑like text. These models can process vast amounts of information to engage in conversations, answer complex questions, and perform a wide range of language‑based tasks. You may have already heard of some like ChatGPT, Gemini, or Claude.
The University of Florida uses NaviGator—a secure, self‑service platform—where students, faculty, and staff can integrate with various LLMs using data in a private environment. The power of these tools extends from our personal lives to our professional responsibilities.
For instance, you can use an LLM to quickly summarize a lengthy research article before your morning rounds, or to brainstorm talking points for a patient education session.
On a personal note, it can help draft a tricky email or even plan a family vacation. As LLMs become more common, knowing how to interact with them effectively is a crucial skill. Learning the right techniques to guide the model helps ensure the results you receive are accurate, relevant, and genuinely useful.
This knowledge can transform a novel technology into a powerful tool that enhances your efficiency and simplifies complex tasks. This series of videos will show you how to harness the power of NaviGator to streamline your workflow.
Prompt Engineering: The CREATE Formula
For more nuanced clinical questions, the CREATE formula offers even greater control over the AI’s output. This video details the six components—Character, Request, Examples, Adjustments, Type of output, and Evaluation—that allow you to build highly specific prompts. You’ll learn how to add critical context, set constraints (like patient allergies or local resistance patterns), and provide examples to guide the model, making NaviGator a more sophisticated partner for complex, multi-step clinical reasoning.
Transcript: Prompt Engineering: The CREATE Formula
The essence of prompt engineering is to design your instructions carefully in order to get precise, useful LLM outputs on the first try. Let’s explore a more advanced method for complex clinical queries, the CREATE formula. This technique gives you even greater control when the details matter most.
CREATE stands for character, request, examples, adjustments, type of output, and evaluation and steps. Character and request are like the role and task in RTF. You’re defining the AI’s persona and what you need it to do, but CREATE adds more layers.
With examples, you can guide the model by showing it what you have in mind, such as considering specific diagnoses or radiographic modalities to compare. Adjustments are where you set the guardrails. You can specify that a patient has a particular allergy, that a recommendation must align with current clinical practice guidelines, or that a previous medication failed.
Type of outputs is the format, just like in RTF. Finally, evaluation, and steps provides the essential clinical context. The patient’s history, relevant lab values, or the urgency of the situation.
Let’s compare. A basic prompt might be, what are the treatment options for pyelonephritis? Using CREATE, a more developed prompt might be character.
You are an attending emergency physician, request, propose an evidence-based antibiotic plan for a 45-year-old female with uncomplicated pyelonephritis. Examples, consider fluoroquinolones and cephalosporins. Adjustments, the patient has a penicillin allergy and local resistance to bacterium is high.
Type of output, a numbered list of options with dosing. Evaluation, she is stable for outpatient management.
The CREATE formula empowers you to handle nuanced, multi-step clinical questions with precision, making each of the LLMs in navigator much more sophisticated partners in your workflow.
HIPAA Compliance with NaviGator
This essential video reviews your legal and ethical obligations under the Health Insurance Portability and Accountability Act (HIPAA). It explains that you must not input Protected Health Information (PHI) unless using a model specifically approved for restricted data. The video also details the university’s data classification system (Open, Sensitive, and Restricted) and clarifies your responsibility to match your data type to an approved LLM on the NaviGator platform before proceeding.
Transcript: HIPPA Compliance with NaviGator
As a clinician, your commitment to patient privacy is paramount. The Health Insurance Portability and Accountability Act, or HIPAA, provides the national standard for protecting patient health information. This legal and ethical obligation remains your highest priority when using any digital tool, including NaviGatorAI.
Critically, this means you should be very careful about inputting protected health information or PHI in NaviGator chat. HIPAA does not apply to de‑identified data, but any information that could be used to identify a patient must be handled carefully. For these types of situations, choose an LLM in NaviGator that is approved to work with restricted data such as llama.
To help you navigate this, the university classifies data into three categories. Open, sensitive, and restricted. Each LLM on NaviGator is designed to work with one or more of these types of data.
Open data is public information that can be shared freely like course catalogs or job announcements. Sensitive data, if disclosed, could harm the university’s functions or reputation. This includes things like research work in progress or internal financial reports.
Restricted data is protected by the law. This category includes all PHI, such as medical records, patient names, Social Security numbers, and any other unique identifiers. Before using NaviGator, you must determine your data’s classification and ensure the model you choose is approved for that level of sensitivity.
For your work, this almost always means ensuring your prompts contain no sensitive or restricted information. If you are ever unsure about your data’s classification, contact your designated data trustee or the UF privacy office before proceeding. Remember, all use of NaviGatorAI must comply with all university policies, including those governing data security, intellectual property, and academic integrity.
Choosing the Right Tool: A Guide to NaviGator Models
Not all AI models are the same. NaviGator provides access to different LLMs, each with unique strengths. This table will help you select the best model category for your specific clinical task, balancing capability with security requirements.
| Top-Tier Models | Fast / Efficient Models | Sensitive Data Models | |
|---|---|---|---|
| Representative Models |
|
|
|
| Key Benefits |
|
|
|
| Key Drawbacks | Not approved for Sensitive / Restricted data (no PHI — UF does NOT host these) |
Less depth / nuance than thinking models |
May trail top cloud models on open-ended reasoning |
| When to Use (Clinical Scenarios) |
|
|
|
NaviGator AI Models are constantly updating. For the most up-to-date information on available models and their approved data classification, visit the NaviGator AI overview page.