Verily’s AI Scientists Jong Ha Lee and Eric Yang presented our paper, The Geometry of Queries: Query-Based Innovations in Retrieval-Augmented Generation (RAG) for Healthcare Question-Answering (QA), at the Machine Learning for Healthcare (MLHC) Conference. The paper introduces a new framework that enhances the accuracy and reliability of RAG systems in healthcare QA. Instead of directly matching user questions to documents, query-based RAG pre-aligns queries with a curated database of answerable questions derived from healthcare content. The acceptance of the paper at MLHC validates our innovative approach to a critical problem. Traditional RAG systems often face a "retrieval challenge" due to the semantic gap between how users ask questions and how information is stored. Our LLM-based filtering mechanism automatically creates a high-quality, query-centric knowledge base, leading to superior performance. This enables building trustworthy digital health applications with appropriate grounding.
Imagine healthcare as unique as you are, tailored precisely to your individual needs. “N of 1” is a documentary series that highlights the promise of precision health to transform lives by harnessing diverse forms of health data and turning them into insights that lead to action — making healthcare more personal and precise.
Better nutrition is key — but hard — for people managing cardiometabolic conditions. Verily’s latest research reveals a new, evidence-based 3Ps approach: ✅ Practical — Be actionable and easy Feedback and guidance must be achievable and low-friction. Participants responded best to small, actionable suggestions — like healthy food swaps, portion tips, or specific meal ideas. The goal is to fit healthy habits into real, feasible routines. ✅ Positive — Celebrate the wins Participants were more motivated when feedback celebrated small wins rather than calling out "bad" choices. Negative words, red symbols, or shaming were demotivating. Starting with encouragement helped participants feel seen and supported. ✅ Personalized — Reflect the individual Meal feedback is most effective when it reflects an individual’s goals, preferences, and real-life context. That means going beyond macros or calories — and asking open-ended questions that invite reflection and autonomy.
Can AI profoundly improve human health? The primary barrier to AI-driven innovation in healthcare is the inability to connect siloed, unstructured data that is organized on different standards, making it difficult for healthcare companies to model on the right data. The Verily platform harmonizes many healthcare datasets — from labs and EHR to unstructured physician notes, genomics and more, with the tools companies need to model and deploy insights into research and care.
There is more health data than ever. But the ability of life sciences and healthcare organizations to govern, analyze, and partner on data remains a challenge, which is imperative for biomedical research and development. Verily Workbench provides the scalable infrastructure and collaborative tools to unlock the value of multimodal health data at scale, across genomics, wearables, imaging, and EHR data.
Hallucinations, outdated information, and accuracy issues are challenges to deploying Large Language Models (LLMs) in healthcare question-answering. How can we build trust? Our new paper introducing Query-Based Retrieval Augmented Generation offers a framework that improves both retrieval effectiveness and the quality of generated responses. Verily Data Scientist @Eric Yang will present the solution and key takeaways (listed below) at the Machine Learning for Healthcare (MLHC) Conference next week from August 15-16. -> LLMs enable automated knowledge base enhancements for retrieval augmented generation (RAG) applications -> Offline alignment enhances performance and reduces inference-time latency -> Comprehensive evaluation is crucial for healthcare question-answering
AI-driven technologies are reshaping healthcare delivery to make it more personalized for you. Help us bring this important conversation to the SXSW 2026 stage by voting for our proposals through the PanelPicker process. ✅ Tomorrow’s Health System: How AI Agents Will Redefine Care - featuring Andrew Trister, MD PhD (Verily), Brian Anderson, MD (CHAI), Brendan O'Leary (former FDA), Riley Ray Griffin (Bloomberg): https://veri.ly/4lkiVCl ✅ How AI is Enabling A Radical Shift in Understanding You - featuring Vindell Washington (Verily), Dr. Geeta Nayyar, MD, MBA (Clinical advisor), John Gerzema (Harris Insights), Alice Park (TIME): https://veri.ly/4murDyX ✅ Rethinking Personalized Health in a GLP-1 World - featuring Carolyn Bradner Jasik, MD (Verily), Caroline Susie RDN, LD (MercerWell), Dr. Phillip Levy (Wayne State U.), Arundhati Parmar (MedCity News): https://veri.ly/3H7GVut
🌟 A Brighter Way to Care 🌟 Watch the full Verily N of 1 episode to see how AI helps care feel more human.
In the first edition of the Verily Tech Blog, we explain the rationale behind using FHIRPath, which is to ensure everyone can speak the same healthcare data language. But getting it to work in code isn’t exactly plug-and-play. The first attempt was messy and labor-intensive. Read our next blog to learn how product engineering made FHIRPath parsing cleaner, faster, and much easier to extend by using ANTLR, a tool that turns a language’s grammar into working code.
How to make your health data work smarter? 🤔 The Verily platform enables modeling and decisioning across multiple health data resources for personal, precise, AI-driven care experiences and workflows
