Photo: Photo: Laurence Dutton/Getty Images
For health IT leaders seeking to deploy large language models in clinical environments, a workshop next month at the 2026 HIMSS Global Health Conference & Exhibition in Las Vegas offers useful perspective.
Attendees of Jeremy Harper's "Deploying Large Language Models for Clinical Workflows" will get an up-close look at building workflows, implementing human-in-the-loop safety, and adapting IT teams to the new artificial intelligence engineering paradigm that's transforming healthcare delivery.
Harper, chief information officer at Owl Health Works, author, and board member of Indiana HIMSS, will guide attendees through LLM best practices for reshaping real-world clinical workflows and optimizing tasks.
His talk will combine hands-on experience and insights into how LLMs work and why they can handle vast amounts of clinical text more effectively than previous AI tools.
Understanding the new landscape
The session is part of a series that's new to this year's global conference: four focused, interactive sessions that combine expert insight with discussion and practical application.
Harper told Healthcare IT News that his workshop – distilled from its full half-day version into 60 minutes – is designed to walk attendees through the critical realities of AI deployments.
He'll demo a sandbox approach using Google AI Studio to convert free-text clinical notes into structured, actionable data. Attendees will follow along and learn to apply a retrieval-augmented generation framework to improve LLM accuracy and advanced prompt engineering to convert unstructured notes into standardized fast healthcare information resources.
While Harper said that he will use a free studio in Google to access a million-token context window, attendees can use an LLM they have access to go through three "very small, very lightweight projects" as a group.
Paradigm shift in clinical engineering
"It's so easy to have little silos of knowledge," Harper said. "Most of us have logged into ChatGPT and done this, that or the other thing with it."
However, he added, "We don't necessarily have that holistic viewpoint: 'How do I get this up and running? How do I make sure that it's functioning with value for my organization?'"
Part of that discussion will be how IT team structures must change, Harper explained. With LLMs, someone with little specialized knowledge can now accomplish tasks that previously took months of study.
"It's an amazing tool" for taking free text, such as in clinical notes, and moving it at scale into discrete data that's actionable.
But at a fundamental level, trying to understand how to design a high-level architecture for an electronic healthcare record-integrated LLM agent requires specifying data flow, security controls and successful key performance indicators for clinical adoption.
Tethering project lifecycles to reality
A core tenet of Harper's approach is engineering for "cost, speed [and] precision," rather than just "power," he explained.
If a healthcare provider is using the most powerful, expensive AI models to process data for hundreds of patients per hour, after all, those costs become prohibitive.
The "smartest" models are the slowest ones and will require the most compute power, he said.
"A more intelligent model is going to have 10 times the cost of a more basic model," Harper noted. "We're looking for the very cheapest model that we can get away with using that still delivers what we need [99% confidence]."
The workshop will also give attendees the opportunity to evaluate LLM outputs for accuracy, bias and patient-safety risk using quantitative metrics and a human-in-the-loop review "kill-switch" checklist.
As part of that discussion, Harper said he will talk through "the mindset" for evaluating an AI project's performance. His framework for screening projects helps determine when AI is a powerful accelerator, or not.
That framework asks, "Where can we use large language models to facilitate and speed up a process, and where is it just mildly hopeless?" he said.
"At the end of the day, [LLMs] can speed up a lot of processes, but there's still that it's not [AI]; it's a next token predictor."
Harper, who recently published a book, Large Language Models for Healthcare: A Practical Guide to Their Process and Evolution, will also answer questions that hit on real-world examples of deploying LLMs into clinical workflows, he said.
"When people get in the room and start talking through and thinking through the reality [of LLMs], a lot of the value of these sessions is what questions come up."
Harper's session, "Deploying Large Language Models for Clinical Workflows," is scheduled for Wednesday, March 11, from 9:45 a.m.-10:45 a.m., in San Polo 3501A, Level 3 at the Venetian at HIMSS26 in Las Vegas.
Andrea Fox is senior editor of Healthcare IT News.
Email: afox@himss.org
Healthcare IT News is a HIMSS Media publication.
