Innovaccer partners with NVIDIA to advance multimodal AI in healthcare
Innovaccer has teamed up with NVIDIA to bring multimodal AI into daily clinical workflows. The collaboration plugs NVIDIA's full-stack platform-NeMo Guardrails, NeMo Framework, Riva Parakeet NIM, Triton Inference Server, and TensorRT-LLM-into Innovaccer's products and data platform.
Deployed across GPU-based AWS and virtual private cloud environments, the setup focuses on safety, scale, and speed. The goal is simple: turn speech, text, and clinical data into real-time support for care teams and operations.
What's in the stack
- NVIDIA NeMo Framework and Guardrails for building domain-specific models with policy-controlled outputs (learn more).
- Riva Parakeet NIM for streaming speech recognition and synthesis built for production.
- Triton Inference Server for serving models efficiently at scale (product page).
- TensorRT-LLM for low-latency large language model inference.
Three focus areas
- Audio AI: Real-time speech recognition to support tools like Access Center Copilot.
- OCR and document understanding: Turning scans and forms into structured notes via InNote.
- Model training and optimization: Using NVIDIA GPUs to improve accuracy, latency, and throughput for Innovaccer's proprietary models.
Why this matters to care teams and IT
- Less manual work: Faster note generation and fewer copy-paste loops from scanned documents.
- Better patient access: Real-time call transcription and guidance to reduce handle time and improve routing.
- Safety by design: Guardrails to control outputs, auditability, and policy enforcement for clinical use.
- Operational scale: Triton and TensorRT-LLM keep latency predictable under peak loads in AWS and VPC environments.
Agents of Care
The partnership also advances Innovaccer's Agents of Care-AI-driven tools that automate repetitive, low-value tasks. Expect targeted support across documentation, routing, intake, and follow-up so clinicians can stay focused on patients.
Abhinav Shashank, Co-founder and CEO of Innovaccer: "Healthcare is entering an era where AI must be able to listen, interpret, and act with precision to deliver better outcomes. Our collaboration with NVIDIA strengthens our ability to build responsible, domain-specific AI, which, when combined with our unified healthcare data platform, can drive meaningful improvements in outcomes and efficiencies."
Trent Norris, Global Head of Business Development for Digital Health at NVIDIA: "Healthcare data is inherently multimodal and deeply contextual. By adopting NVIDIA's acceleration libraries and platform, speech frameworks, and pre-trained models, Innovaccer is standing up production-ready systems that reason across speech, text, and clinical data in real time. This isn't just experimentation-it's healthcare AI being operationalized at scale."
How to act on this
- Pick 1-2 clear use cases: Access center calls, intake notes, discharge summaries.
- Define success: Handle time, first-call resolution, note completion time, note quality, clinician satisfaction.
- Set guardrails: PHI handling, consent, logging, prompt policies, human-in-the-loop review.
- Plan integrations: FHIR/HL7, EHR sandbox testing, SSO, device microphones for speech.
- Roll out in stages: Shadow mode → limited pilot → broader deployment with weekly KPI reviews.
What to watch
- Speech accuracy on noisy lines and accents; on-device vs. cloud inference policies.
- OCR performance on legacy scans and faxed forms.
- Latency and costs under peak volumes; autoscaling behavior.
- Documented reductions in burnout and improved access metrics.
If your organization is planning team upskilling around AI workflows and agentic tools, explore practical training by role (courses by job).
Your membership also unlocks:
