Research
The evidence behind AI that can't afford to fail
Published methods and open benchmarks, proven on millions of real clinical interactions.
Model reasoning
General purpose LLMs excel at solving math exams but struggle with complicated real-world healthcare reasoning. Learning to reason from complex healthcare histories, recent medical research, and laboratory results requires a specialized reasoning agent with embedded knowledge of millions of related healthcare cases.
Policy optimization
Policies define the protocol followed by a machine learning model and are learned from human guidance. General purpose LLM policies are defined by non-healthcare-trained individuals, but optimizing policies through feedback from healthcare professionals enables learning protocol from the best in class.
Alignment & interoperability
AI systems are inherently bad at knowing when they don't know. Building interpretability into AI allows healthcare professionals to understand the reasoning process and locate the source of its output, while alignment AI models interpret the output of other AI models at scale, flagging inconsistencies before they harm.
- Symphony for Speech-to-Text: Supporting Real-Time Medical Voice Interfaces
- Symphony for Medical Coding: A Next-Generation Agentic System for Scalable and Explainable Medical Coding
- A medical coding language model trained on clinical narratives from a population-wide cohort of 1.8 million patients
- GIM: Improved Interpretability for Large Language Models
- A Text-To-Text Alignment Algorithm for Better Evaluation of Modern Speech Recognition Systems
- Code Like Humans: A Multi-Agent Solution for Medical Coding
- Voice as a Health Indicator: The Use of Sound Analysis and AI for Monitoring Respiratory Function
- FactsR: A Safer Method for Producing High Quality Healthcare Documentation
- Normalized AOPC: Fixing Misleading Faithfulness Metrics for Feature Attribution Explainability
- An Unsupervised Approach to Achieve Supervised-Level Explainability in Healthcare Records
- A Retrospective Study on Machine Learning-Assisted Stroke Recognition for Medical Helpline Calls
- MSViT: Dynamic Mixed-Scale Tokenization for Vision Transformers
- Automated medical coding on MIMIC-III and MIMIC-IV: A Critical Review and Replicability Study
- Self-Supervised Speech Representation Learning: A Review
- Model-agnostic out-of-distribution detection using combined statistical tests
- A Brief Overview of Unsupervised Neural Speech Representation Learning
- Benchmarking Generative Latent Variable Models for Speech
- On scaling contrastive representations for low-resource speech recognition
- Do We Still Need Automatic Speech Recognition for Spoken Language Understanding?
- Do end-to-end speech recognition models care about context?
- Hierarchical VAEs Know What They Don't Know
- MultiQT: Multimodal Learning for Real-Time Question Tracking in Speech
- BIVA: A very deep hierarchy of latent variables for generative modeling
- On the Inductive Bias of Word-Character-Level Multi-Task Learning for Speech Recognition
- Utilizing Domain Knowledge in End-to-End Audio Processing
- Semi-supervised generation with cluster-aware generative models
- Auxiliary deep generative models
- Ladder Variational Autoencoders
- Recurrent spatial transformer networks
No research items match your filters
Try adjusting your search or filters to find what you're looking for.
Join our research team
Build frontier AI for industries that can't afford to fail. We are always looking for brilliant minds dedicated to innovating in critical industries.
View open positions