Healthcare AI in the Real World: From Genomics to Governed Graphs

🚀 Join Us for DataPhilly’s Tech Talks at Perpay! 🚀
This event is held in partnership with Perpay, a company dedicated to building simple and inclusive financial products that help members create healthy habits and achieve economic stability.
We’re also proud to be supported by our Gold Sponsor, Liberty Personnel Services, widely recognized as one of the region’s leading direct placement and contract recruiting firms.

Event Schedule:
Doors opens at 6:00 pm ET
6:00 - 6:30: Event start and networking, DataPhilly intro
6:30 - 7:15: Jim Havrilla, Survivor Bias: My Career in Healthcare Data Science, followed by Q&A
7:15 - 8:00:Thomas Elliott, Building Trust with Constrained Graph Reasoning at Biosole International Group, followed by Q&A
After 8:00: Networking time

Speakers:
Jim Havrilla, PhD, Senior Director of Data and AI at Pathnostics (jimhavrilla.github.io): Survivor Bias: My Career in Healthcare Data Science
Abstract: Healthcare data will humble you. Over the past decade I've worked with population-scale genomic variation, rare disease phenotypes, tumor sequencing pipelines, clinical notes, and scattered unstructured data from a diagnostics laboratory—and at every stage, the data found creative new ways to be weird. This talk is a practitioner's tour through that journey, with lessons from each stop: building a map of evolutionary constraint across the human genome (published in Nature Genetics); extracting clinical phenotypes from the chaos of electronic health records to prioritize genes for rare diseases; wrestling with the structural variants that refuse to fit neat categories in cancer sequencing; and now, at a laboratory diagnostics company, discovering just how strange real-world operational data can get—and building AI-powered decision support tools for pathologists despite it. This talk focuses on patterns and accomplishments, not a deep dive into any single method. Which assumptions survive contact with clinical reality, which ones fail, and what has a decade of being surprised by data taught me about making data genuinely useful in healthcare? Ultimately, it’s a talk about how to build systems that help people, even when the data refuses to behave.
Bio: Jim Havrilla, PhD, is a healthcare data scientist and engineering leader with a decade of experience turning messy biomedical data into tools that work. His PhD research at the University of Utah produced the Constrained Coding Regions (CCR) model — a map of evolutionary constraint in the human genome was published as the cover article for Nature Genetics (Jan 2019). At the Children's Hospital of Philadelphia, he led development of clinical informatics tools including Phen2Gene (a phenotype-driven gene prioritization tool), PhenCards (a phenotype knowledge base he continues to maintain), and Termviewer (NLP tool for curating and extracting clinical phenotypes from electronic health records). In industry, he has built enterprise data architectures and analytics platforms from the ground up and developed AI-powered clinical decision support tools for diagnostics — work featured by Databricks and covered by Conexiant Pathology. He believes the best technology leaders stay connected to the code and data that make strategy real.

Thomas Elliott, Head of AI and ML Systems @ Bio-Sole International, Founder of Augmented Intelligence Advisory Group (https://youraugmentedlife.com/): Building Trust with Constrained Graph Reasoning at Biosole International Group.
Abstract: Healthcare AI breaks down when outputs can’t be traced back to governed data and a defensible reasoning path. This talk focuses on some of the practical rules for building trustworthy systems where deterministic ML-derived signals become evidence inputs to a versioned knowledge graph and policy-driven traversal. This provides a system that can explain calculated results and how a specific conclusion was reached through a constrained, reviewable path. This aligns with my core axioms for AI in production: building functions that are inspectable, auditable, explainable, supportable, and scalable to augment human capability.
We’ll cover how to bind datasets, model versions, and graph/ontology versions into a single reproducible release artifact, and how to produce deterministic “patient graph” exports that can be replayed for audit, QA, and clinical review. You’ll see patterns for evidence preservation, data contracts, traversal constraints, and causal traceability that link every decision to the exact inputs, transformations, and graph paths used. ...and how transactions are exportable with the data used in summarization to ensure that humans in the loop have auditable sources to support explanations.
Bio: Thomas Elliott is an AI and distributed data systems leader focused on building trustworthy ML in production. He leads AI and ML systems at 'Bio-Sole.ai', where teams turn real-world sensor data into measurable outcomes using governed pipelines and causal graph reasoning. His background spans telecom at Ericsson and Amdocs, cybersecurity at Entrust, and product startups, including Bridgewater Systems and CENX in Ottawa, Canada.
Please note, by RSVPing this event you agree to our Code of Conduct
Looking forward to seeing you there! 🚀