Redact PHI on the request path
Configured PHI detectors identify sensitive fields and the REDACT action removes them before the governed request reaches a model.
HealthYoda · Customer storyProtect PHI before healthcare AI agents call a model.
HealthYoda puts an AI workforce in clinics. DVARA is the AI gateway in front of its agents and models, where configured controls redact protected health information and eligible repeated requests can be served from the semantic cache.
The healthcare workflow
An AI workforce connected to clinic operations.
HealthYoda provides AI agents for patient intake, clinical documentation, scheduling, coding, billing, follow-up and other EHR-connected work.
Those workflows can combine long clinical context with PHI. HealthYoda uses DVARA as the governed path between its agents and configured models, so privacy and cost controls do not have to be rebuilt inside every agent.
The governed workflow
One control point in front of agents and models.
HealthYoda combines request preparation with DVARA’s configured privacy and cache controls without changing its customer-facing workflow.
Compress long clinical context
HealthYoda’s request-preparation step reduces the model input carried by long clinical context before the request enters the governed model path.
Cache only eligible requests
Caching is limited to requests HealthYoda marks as safe and repeatable. Patient-specific note generation and PHI-bearing requests stay outside this model.
Illustrative workload model
What the token economics can look like.
These figures model a representative scribe workload. They are not audited HealthYoda production results and should not be read as a customer savings claim.
10,000Monthly scribe requests
28%Modelled input-token reduction
22MModelled input tokens avoided
18%Eligible non-PHI cache hits
Before compression7,800 tokens/request
After compression5,600 tokens/request
Across 10,000 requests22M input tokens avoided
Cache hits are modelled only for eligible, repeatable non-PHI requests. Patient-specific clinical-note generation is excluded. Actual results depend on prompt shape, model, cache policy and traffic mix.
The compliance boundary
A control for HIPAA workflows—not a certification.
Configured PHI redaction helps HealthYoda support its HIPAA compliance controls. DVARA does not make an application HIPAA compliant by itself; deployment, access controls, data handling, vendor agreements and operating procedures still matter.
See how DVARA redacts sensitive data →Your healthcare workflow
Bring one agent and one PHI boundary.
Map where DVARA should redact, cache and record evidence, and where prompts can be compressed before the workflow reaches production.
Book a healthcare AI working session