AI at the Point of Care: How Real-Time Clinical Knowledge Tools May Transform Antimicrobial Stewardship

This is the latest episode of our From Pathogen to Infectious Disease Diagnosis podcast, where we discuss the relationship between clinicians and laboratory professionals and detail the latest in diagnostics.

Artificial intelligence (AI) is increasingly reshaping clinical decision-making, and a new generation of point-of-care tools is targeting one of health care’s most persistent challenges: antimicrobial stewardship. In a recent podcast discussion, Eolas Medical Founder and CEO Declan Kelly, MBBS, outlined how limited access to localized guidelines has historically hindered optimal prescribing. As a former emergency department physician, Kelly understands that clinicians often face time-sensitive decisions while navigating fragmented systems, outdated intranets, or hard-to-find protocols, particularly in high-pressure environments such as the emergency department.

Kelly explained that the core issue lies not in the absence of guidance but in its accessibility at the moment it is needed most. “It was very hard to access when you were making decisions at the point of care,” Kelly said.

Kelly said his AI-driven software company has created a clinical knowledge platform, Ask Eolas. The platform addresses this gap by aggregating institution-specific protocols and enabling clinicians to query them instantly using natural language. By prioritizing local antimicrobial resistance patterns and hospital-specific policies, the tool ensures that recommendations are both relevant and actionable, rather than relying on generalized or external sources.

Beyond convenience, the implications for patient safety are significant. Early evidence suggests that AI-assisted tools can dramatically reduce prescribing errors, particularly compared with traditional PDF-based or static guideline systems.

“There was 100% prescribing safety…0 errors made when the AI was used,” Kelly said, referencing a recent study evaluating their platform in simulated clinical settings.

Kelly emphasized that as AI adoption accelerates across health care, responsible implementation remains critical. Although the technology offers faster decision-making and improved guideline compliance, it must be paired with rigorous validation, clinician oversight, and transparent sourcing. With the ability to integrate into electronic health records and even anticipate clinical needs based on patient data, such tools signal a shift toward more proactive, data-driven care, where the right answer is not just available but delivered exactly when and where it is needed.

In thinking about oversight, Kelly explained that there are processes in place.

“What we do is a 2-pronged approach, and one is we get a subset of them actually reviewed by clinicians. And two is that every single one of them goes through an automated second-judge system. So there's a second large language model as a judge that says, ‘Hey, here was the answer that was given. Here are the sources that supplied that answer—review this answer as well and score it to see if this was accurate and safe,’” Kelly said. “Then we get a subset of those, because you obviously can't get millions reviewed manually, but we get a subset of those then reviewed by physicians as well, and we can see every single question that was asked, and we score that and review that from an observability point of view, because at the end of the day, their AI is probabilistic. It's a statistical model, so it can never be 100%. The world's most powerful AI will always be 99.999%.”