{"slug":"mendel-accelerates-clinical-trial-matching-with-databricks","url":"https://findausecase.com/use-cases/mendel-accelerates-clinical-trial-matching-with-databricks","title":"Mendel accelerates clinical trial matching with Databricks","description":"Healthcare AI company Mendel used Databricks to pretrain and fine-tune custom large language models like Llama 3.1 on around 200 billion proprietary tokens for its HyperCube clinical data platform, reducing model training time from three months to one month while maintaining HIPAA compliance.","company":"Mendel","industry":"Healthcare","aiCapabilities":["Large Language Models","Generative AI","AI Model Development & MLOps"],"technology":["Llama 3.1","Agent Bricks","Unity Catalog"],"deployment":"Public Cloud","problemStatement":"Fragmented clinical data across systems like EMRs and different health networks made it difficult for Mendel's HyperCube users to perform real-time queries or extract actionable insights, and building the disease-specific ontologies needed for semantic querying was time-consuming and resource-intensive.","solutionApproach":"Mendel used Databricks to pretrain and fine-tune custom large language models like Llama 3.1, starting from a partially trained model, continuing training on public datasets, and then integrating around 200 billion proprietary tokens. Databricks also enabled Mendel to unify its data and ETL pipelines and run federated queries across multiple data sources.","businessValue":"Databricks helped Mendel accelerate model training, reducing the timeline from three months to just one, while ensuring HIPAA compliance and enhancing the accuracy of their AI solutions.","evidence":{"band":"high"},"sourceUrl":"https://www.databricks.com/customers/mendel","dates":{"publishedAt":"2026-09-11T09:03:50.741Z","publishedAtSource":"pipeline","updatedAt":"2026-09-11T09:03:50.741Z"},"license":"Open for reading and citing with a link to https://findausecase.com/use-cases/mendel-accelerates-clinical-trial-matching-with-databricks. Bulk republication requires permission."}