The Technology Backbone for Federated Research Networks and Privacy-First Clinical AI
Virtual Pooling™ connects health systems into federated networks. Hummingbird™ turns clinical notes into structured data privately and at scale. One technology backbone built to overcome the privacy constraints that have long limited discovery.
Virtual Pooling™: The Technology Platform for Building and Operationalizing Federated Networks.
End-to-end automation of federated research, with built-in data privacy.
Easy to use
Federated Data Harmonization Capability
Readily Available Federated Biostatistical Operations
Built-In Privacy, Security, Access Control, and Auditability
See what the the platform solves
Hummingbird™ - Unlock the Evidence Buried in Clinical Notes
Extract research ready variables from unstructured sources – efficiently, securely, and at scale.
Accuracy That Generalizes
--------
01
Hummingbird™ matches LLM-level accuracy and F1 performance using proven clinical NLP methods, validated in peer-reviewed research across use cases.
Cost Efficient Extraction--------
02
Built for standard compute environments, Hummingbird™ turns unstructured clinical notes into structured data without relying on GPUs or external processing power.
Privacy by Design, Deployed Anywhere
03
Hummingbird™ runs on CPUs only and never contacts frontier LLMs, keeping institutions in control of their data wherever it is deployed.
Our Technology at Work
Explore how our technology helps researchers overcome privacy constraints to answer their scientific questions.
Low-cost Feature Extraction from Clinical Notes Using Hummingbird: ECOG Performance Status of Lung Cancer Patients
Virtual Pooling is Accurate and Lightweight for Multi-Institution Causal Inference without Centralized Data Sharing
In collaboration with
UCSF · Genentech · UCI
Extraction of Complex IBD Features from Clinical Notes Using Hummingbird: A Cost-Effective, Privacy-Preserving Alternative to Large Language Models
Use of a Small Language Model to Identify MG-ADL Scores from Encounter Notes in an EMR System
In collaboration with
Landmark Science · UCSF
Virtual Pooling: Accurate, Scalable, and Privacy-preserving Analytics Without Aggregating Patient Data
A Secure Method for Fine-Tuning and Using Clinical Language Models Across Multi-Center Datasets
In collaboration with
University of Alabama · UCSF