Damon Ragusa

At Sohar, we continually explore new ways to leverage technology to make healthcare more equitable. Most recently, we’ve added large language models (LLMs) into our eligibility workflow. Machine learning has always helped us process structured data at scale; however, critical details can get lost in free-text, and LLMs are helping us bridge that gap. Accuracy rates are increasing from ~80% to ~95%, especially with smaller payers, which results in fewer manual calls for staff and faster, more reliable answers for patients. LLMs are another step toward building healthcare infrastructure that works better for everyone. #HealthcareInnovation #DigitalHealth #AI #SoharHealth #Leadership

17

2 Comments

ICYMI: NVIDIA is betting big in healthcare. And one key company that it has chosen to work with: Verily— a titan in the field of precision medicine. Why this partnership, why now? Because the time has never been better with how deeply AI can make sense of large swaths of data to predict disease earlier and personalize medicine. This is the science and toolset that Verily is working on. This is huge news for the life-sciences world. Read more in the press release below. CC: Janette Ciborowski, Stephen Gillett, Steven Cooper, Trent Norris, Myoung Cha

183

2 Comments

[#RecommendedRead] As generative AI becomes embedded in how we search, learn, and make decisions, we’re entering a new era of information risk—not just misinformation, but an AI-driven infodemic. This short but powerful piece from Nicolas Chehade MPsych, MGA/MPP at ThriveSpace Collective, “A Clinical Framework for Navigating the Generative AI Infodemic,” offers a compelling way forward. Rather than reacting with fear or blind adoption, it proposes a clinical, diagnostic approach to AI-generated information—one that emphasizes discernment, human judgment, and systems-level accountability. What I appreciated most is how it reframes the challenge. This isn’t about rejecting AI, but about building the muscles to question, verify, and contextualize what AI produces, especially in high-stakes domains like health, policy, leadership, and education. For anyone grappling with how to use AI responsibly—without outsourcing critical thinking—this is well worth your time. https://lnkd.in/eghtdKay 👉 Curious how others are navigating trust, judgment, and responsibility in an AI-saturated information environment? #ResponsibleAI #GenerativeAI #DigitalTrust #CriticalThinking #HealthAndTechnology #Leadership #FutureOfWork #HumanCenteredAI

9

2 Comments

🎧 New clip from our X Spaces with @SuperRobotGX! In this segment, we explain: • How our proprietary AI actually works inside our DSD model • Why operational AI (not hype) creates real competitive edge • How ASC 606 workflows keep us fully compliant as we scale GPOX is building AI that makes distribution faster, smarter, and more reliable. Watch the quick breakdown 🎥👇

6

Clinicians want to use AI but don’t have the bandwidth to evaluate all the different models. Meet Health Olymp, an AI tool purpose-built for clinicians. Here’s why this matters 👇 1. Health and life science experts gain instant access to frontier models for free. No procurement hoops or IT bottlenecks. Every clinician gets their own sandbox to explore. 2. Clinicians can compare models side by side and pick the one that works best for their use case. That transparency builds trust and sparks ideas for novel AI applications in care delivery, diagnostics, and research. 3. Most AI benchmarks are too generic or disconnected from clinical reality. Health Olymp brings frontline feedback into the loop, helping the industry identify models that perform reliably in real world clinical environments. Congrats to Ruslan Nazarenko and the Lumos team as they put AI model benchmarking directly into the hands of healthcare professionals. ⬇️ (link in comments)

148

53 Comments

Benchmarking the Future of Medical AI Agents: Really interesting paper introducing the first benchmark to evaluate how large language models function as medical agents inside EHRs. Traditional LLMs (like ChatGPT) act as chatbots providing answers to prompts. But "agents can proactively interpret high-level instructions, plan actions, interact with external systems, and iteratively refine their responses." That shift could help reduce clinician burnout, improve clinical care quality, and close workforce gaps. MedAgentBench tests 300 clinically derived tasks across 100 patient profiles (700K+ data points) in a FHIR-compliant virtual EHR. State-of-the-art models showed promise but with wide variation. Claude 3.5 Sonnet v2 topped the field at a 69.7% success rate. Impressive, but still far from reliable for clinical use where errors in medical records can have serious consequences. However, if current trends hold (as METR notes, AI’s ability to complete long tasks is doubling roughly every 7 months), today’s ~70% performance could reach clinically sound thresholds much sooner than expected. Just as MedAgentBench creates a virtual EHR to test AI agents in medicine, education needs its own suite of agentic benchmarks. Imagine an assessment agent that continuously measures student progress, an SIS agent (like an educational EHR) that integrates student records, a science of reading agent that adapts instruction to phonics and fluency needs, an HQIM agent that generates and sequences high-quality instructional materials, and a coherence agent that reviews all of this to ensure alignment with learning standards and the evidence base. Benchmarks could test whether these agents work together reliably—retrieving data, planning instruction, scaffolding practice, and adjusting in real time. Without such benchmarks, we risk deploying systems that look impressive in demos but fail to deliver meaningful learning gains. https://lnkd.in/gnzCwbi7

10

1 Comment

There are truck loads of companies making headlines for AI in healthcare and medicine. This one is quietly-- absent the VC-backed overhype-- nailing real value with a workflow app tailored for med students and CME, centered around AI from the apps' original design, not a bolt-on afterthought. Dendritic Health AI "Dendritic Health's Neural Consult AI Achieved 100% Accuracy on the USMLE Step 1, 2, and 3 Benchmarks" https://lnkd.in/epV4GW2b

40

Sophont just released Medmarks v0.1: a new evaluation suite for assessing the medical capabilities of LLMs. This suite is the largest completely open-source automated evaluation suite for medical capabilities, with a total of 20 benchmarks. It covers tasks like answering patient questions to detecting errors in clinical notes. Here are the early results (full results at medmarks.ai). Interesting to filter them by model size, open-source, etc:

23

1 Comment

Are you leveraging AI to enhance your understanding of personas and data? In our latest episode of Answers with Ahava + Friends, we dive into how AI acts as an analyzer, providing valuable feedback. Kelli Buerger, Director of Marketing at UChicago Medicine, shares how her team uses AI for a foundational check and then further validates insights. 🎧 Want the full conversation? Watch it here: https://lnkd.in/ef_hWzDV

19

3 Comments

New Podcast Episode: The AI Blueprint for Precision Medicine We’re excited to share a new episode of the AI ViVE Podcast featuring our Founder & CEO, Don Brown, MD, in conversation with Sandy Vance. In this episode, Don discusses how AI-driven clinical personas are transforming massive biological datasets into actionable insights, enabling precision care that works in the real world. From his entrepreneurial journey to Bioscope AI’s subscription-based model and early clinical impact, the conversation explores how personalized medicine is shifting from theory to practice. Precision medicine isn’t a future concept. It’s happening now, powered by AI. 🎧 Listen to the full episode here: https://hubs.li/Q043P7sW0 #Longevity #Bioscopeai #ConciergeDoctors #PrecisionMedicine #ArtificialIntelligence #HealthcareInnovation #DigitalHealth

21

1 Comment