H2O.ai and UCSF CDHI Interview with Healthcare IT Today at HIMSS 2022 | Document AI
- Generative AI
- H2O Document AI
Overview
John Lynn of Healthcare IT Today interviews Bob Rogers (AI expert in residence, UCSF Center for Digital Health Innovation) and Sri Ambati (H2O.ai CEO) at HIMSS 2022 about UCSF's Document AI referral automation pipeline. UCSF receives 1.4 million faxes a year, roughly 300,000 of them referrals, once processed manually through a chain of review queues. The conversation covers how H2O and UCSF co-created the model despite outside experts calling it unsolvable, and the resulting gains in faster scheduling and fewer errors.
Key takeaways
- How UCSF's Center for Digital Health Innovation used H2O Document AI to automate a referral pipeline handling 1.4 million faxes a year, about 300,000 of them referrals
- Why outside experts initially said the extraction problem couldn't be solved, and how the first working results arrived within three weeks
- How Document AI combines language context with document shape or layout to correctly interpret fields like patient names
- How care coordinator corrections feed back into retraining the model, described as the difference between ongoing machine learning and a static machine-learned system
- The resulting outcomes: faster patient scheduling, fewer errors from manual review variability, and strong adoption from care coordinators
- How the H2O platform is used to continuously benchmark and validate models in production as data drifts
Start owning your AI