Can gen AI bring order to medical records? This startup is giving it a shot.

One of the holy grails of healthcare has long been the realization of comprehensive, accurate, and insightful patient records — a single reliable view of a person’s lifelong medical journey, with broad accessibility but also broad security and privacy. No more sifting through fragmented data, grappling with conflicting information, or relying solely on intuition. No more avoidable issues because an entry is missing or wrong.

Mendel, a San Jose-based startup, believes it has discovered the path to unifying electronic health records: unique AI models that can reason and unlock the vast potential of clinical data warehouses.

Mendel was co-founded by a trained-physician, Karim Galil, and an AI scientist, Wael Salloum, in 2017. After practicing medicine, Karim had a vision to build an AI that can clinically reason with 100% explainability and avoid hallucinations. Today, Mendel is one of the leaders in generative AI for healthcare and life sciences with a leadership team from leading life science institutions and technology companies, blue chip investors, and some of the world’s leading life sciences companies as customers.

Mendel's mission is to move healthcare towards a more objective, data-driven approach, while simultaneously preserving the vital role of physician expertise. They achieve this by structuring the complex world of unstructured clinical data, reconciling conflicting information, and generating comprehensive patient journeys.

These journeys offer a holistic view of a patient's medical history, empowering clinicians with the insights needed to make informed, data-driven decisions. Mendel’s latest offering is Hypercube, a chat interface designed to make accessing and analyzing clinical data in any setting easier and more reliable, and thus facilitating better decision-making and the improved health outcomes that come with it.

Looking for a cloud partner with a similar emphasis on building data-driven insights for its customers, along with a leading role in the ongoing development of AI, Mendel turned to Google Cloud as the platform on which to develop its models and scale its business.

Mendel determined that Google Cloud's extensive computational resources and global reach could not only help the startup with its deployment of clinical AI solutions — it helps satisfy Dr. Galil’s vision to accelerate the adoption of AI in healthcare, with the ultimate goal of improving patient outcomes and revolutionizing the field.

Clinical AI technology reinvents patient records

For more than two decades, healthcare systems have been moving — albeit with some amount of struggle, particularly in the United States — toward greater adoption of electronic health records. Part of the challenge has been just how fragmented and siloed patient care in the U.S., as people bounce between different primary care doctors, specialists, hospitals, and clinics as they move between locations or even jobs that change their care options.

It would take a person a considerable amount of time to collect, combine, and analyze all this data. Mendel greatly accelerates this process by employing a unique blend of technologies to achieve its vision for clinical AI.

Leveraging the power of generative AI, Mendel's AI models learn complex patterns and relationships within patient data. Mendel leverages a clinical hypergraph, a multi-dimensional knowledge graph based on knowledge representations from enriching and structuring medical information and relationships. This approach helps ensure explainability and consistency in AI outputs.

Mendel’s Neuro-Symbolic Reasoning System is built specifically for clinical data workflows and analytics, and couples large language models with a clinical knowledge hypergraph. The AI system is able to leverage the latest technology advances in large language models, while ensuring explainability and limiting hallucinations with Symbolic AI.

These features add up to a system that is readable and reliable to medical practitioners because it’s grounded in the knowledge and shorthand common to any clinic or emergency room.

Once it analyzes a clinical situation — which could be information as diverse as a series of pathology notes or a patient’s social history that could reveal external health risks — the system can not only offer clear recommendations to the clinician so they can make a more informed decision, but also the reasoning behind the decision is transparent and easily queried. This observability is crucial to building greater confidence in practitioners that they can trust the recommendations of products like Hypercube.

In a traditional research setting, to perform clinical data analytics, practitioners are burdened with complex clinical data workflows to process the vast data sets of structured and unstructured data found in healthcare settings. Practitioners have been able to leverage standard, open-source data models like SNOMED CT, which has enabled marginal improvement to clinical data workflows.

The gains have remained limited, though, due to the lack of clinical reasoning — the exact challenge Mendel is solving for.

Mendel has conducted a study comparing its Neuro-Symbolic AI system against traditional SQL queries with SNOMED CT data models (the study was led by a Permira Private Equity-backed non-regulated research firm). This study found that Mendel’s AI system achieved a 10x increase in efficiency, 90% reduction in query error rate, and more than 80% cost reduction compared to traditional workflows with SNOMED CT.

Bridging the Gap Between AI and Human Expertise

Hypercube in particular is meant to bring new levels of clarity to medical data, and enhanced outcomes along with it. Mendel understands the importance of trust in healthcare and the need to balance that against the immense opportunity that technology presents.

The Hypercube platform offers some novel solutions built with the help of Google Cloud’s AI models tuned to the specific industry needs of Mendel’s customers:

• Hypercube Cohorts: Can define cohorts and discover insights leveraging diverse data sources with 5x increase in accuracy and scale.
• Hypercube Charts: Chat with medical records through a user-friendly interface with sophisticated clinical reasoning to complete chart reviews for visits, pre-authorization, trial matching and more with 30%+ more efficiency.
• Hypercube Analyst: Query in a low-code interface powered by Mendel’s Knowledge Hypergraph reasoning engine that captures semantic, temporal and hierarchical relationships, enabling 10x faster querying.
• Hypercube Build-Your-Own (BYO): Customize a tailored Hypercube co-pilot that leverages advanced clinical reasoning to fit your specific data needs & challenges. BYO comes with pre-built copilots for complex tasks like EMR-to-EDC, site activations, commercial alerts, and custom NLPs and LLMs.

The Future of healthcare is data-driven

When you train a new AI model on medical records, it simply reads the text and syntax without understanding whether something is a symptom or a side effect. Mendel has created a clinical AI that can reason and understand the semantics of medical records.

“What sets our system apart is that we took the hard route — we didn’t just adapt existing technology; we built a fundamentally new approach,” Galil says. “By coupling large language models with a clinical hypergraph, we developed a groundbreaking technology tailored to the unique needs of healthcare.”

Mendel AI is leading the charge towards a future where data empowers healthcare professionals to provide the best possible care for their patients. Their innovative approach to AI, combined with the power of Google Cloud technology, is paving the way for a more objective, efficient, and ultimately, human-centered healthcare system.