Trustworthy AI
Trust, transparency and safety are sacrosanct when it comes to patient health. Learn how we're continually developing our medical-grade AI system.
Built for healthcare
Designed for clinical language and real-world workflows
Our medical AI assistant understands medical terminology and operates accurately within the clinical domain. By utilizing training data closely aligned with real-world use cases, our models have deep expertise in healthcare contexts.
Evaluation
Rigorous testing and clinician evaluation
For every new feature or model change, we run human evaluations before launch. Clinicians score both quality and safety (1-5) across the dimensions below.
Readability
Is the summary well-written using professional language as a trained physician?
Completeness
Does the summary include the critical clinical information needed for care?
Curation
Does the summary exclude irrelevant noise and focus on what matters clinically?
Correctness
Does it avoid hallucinations, knowledge gaps, faulty logic, and bias?
We also capture overall usefulness and ethicality scores, and use failures to drive targeted improvements in prompts, retrieval, and model behavior.
Safety partner
Partnered with Guardrails AI
Abstractive Health partnered with Guardrails AI to further bolster the safety and reliability of our AI solution for clinicians. Guardrails offers an additional validation framework that reinforces the integrity of our LLM applications. Through their services, users can confidently utilize Abstractive Health without concerns about patient data breaches, unsettling hallucinations, or system inconsistiencies.
Monitoring
Continuous performance monitoring
Once the AI system is online, we combine automated faithfulness signals with physician review to monitor generated outputs. If scores fall below a threshold for a sustained period, we investigate and take action quickly.
Rubrics
Transparent clinician rating criteria
Each quality dimension has clear 1-5 rubrics (1 = lowest quality, 5 = highest). In addition, we collect overall usefulness and ethicality scores to ensure outputs meet clinician expectations and safety standards.
Publications
The research behind our clinical AI.
Developing and Evaluating Large Language Model-Generated Emergency Medicine Handoff Notes
By Vince Hartman, MS; Xinyuan Zhang, PhD; Ritika Poddar, MS; Matthew McCarty, MD; Alexander Fortenko, MD, MPH; Evan Sholle, MS; Rahul Sharma, MD, MBA; Thomas Campion Jr, PhD; Peter A. D. Steel, MA, MBBS
Dec 03, 2024
A Method to Automate the Discharge Summary Hospital Course for Neurology Patients
By Vince Hartman, MS, Sanika S Bapat, MS, Mark G Weiner, MD, Babak B Navi, MD, Evan T Sholle, MS, Thomas R Campion, Jr, PhD
Aug 28, 2023
A Day-to-Day Approach for Automating the Hospital Course Section of the Discharge Summary
By Vince Hartman, BA and Thomas R. Campion, Jr., PhD
May 23, 2022
Want to see trustworthy clinical AI in action?
Try Abstractive Health and review patient summaries with evidence.
Stay ahead of the curve in healthcare innovation.
Connect
Abstractive Health
Resources
©2026 Abstractive Health. All Rights Reserved.
