Radpair

Modernizing Healthcare with AI: How RADPAIR and Fireworks Unlock Smarter Radiology Workflows

Kimi K3 on Fireworks: Frontier Intelligence You Can Own

Blog

Radpair Modernizing Healthcare with AI: How RADPAIR and Fireworks Unlock Smarter Radiology Workflows

PUBLISHED 11/9/2025

Table of Contents RADPAIR’s Mission: Smarter, Safer Radiology Workflows The Challenge: Fragmented Workflows and Antiquated Standards RADPAIR’s SDK: Enabling Intelligent, Safe AI Workflows Fireworks: Scalable, Low-Latency AI Infrastructure Before & After: The Transformation to AI-Enabled Radiology Workflows From Radiology to Healthcare at Scale: Smarter Workflows Powered by AI About RADPAIR About Fireworks

Table of Contents

Executive Summary RADPAIR is transforming radiology workflows with its intent to create an open-source SDK standard – anchored by the Report Document Schema (RDS) and Actions and Event Protocol (AEP) – which enables safe, intelligent AI interactions across reporting systems. RADPAIR’s SDK establishes a new AI orchestration standard for healthcare, designed for adoption by coalition partners to enable interoperable, safe, multi-agent workflows across institutions. Fireworks provides the enterprise-grade infrastructure and orchestration platform for RADPAIR’s fine-tuned models and multi-agent pipelines, ensuring real-time, scalable, and compliant performance. Today, radiologists at institutions including Radiology Partners, which handles 40-50 million cases annually, benefit from AI-assisted workflows that integrate real-time dictation and generative AI structured reporting, reducing cognitive load, accelerating throughput, and improving diagnostic confidence. Key performance gains observed in production include: • Report turnaround : Reduced from 15-20 seconds → 2-5 seconds , meeting clinical Service Level Agreement (SLA) requirements • Workflow efficiency: Early data suggests ~25% reduction in time per case , thanks to auto-populated fields, smart suggestions, and reduced context switching • Reporting accuracy : Pilot studies and independent research indicate 12% fewer reporting errors ( Radiology Business ). • Scalability : Supports 1,000+ concurrent users, with sub-second Speech-to-Text (STT) latency and near real-time display of intermediate transcripts (<200ms).

By combining RADPAIR’s innovative AI orchestration with Fireworks’ scalable, low-latency infrastructure, the partnership unlocks clinically meaningful impact: radiologists focus on interpretation and patient care, patients receive faster, more accurate diagnoses, and hospitals gain auditable, compliant workflows without building their own AI infrastructure. RADPAIR Radiology Transcription Key Outcomes at a Glance Outcome Before After / Target Report Turnaround 15-20 seconds 2-5 seconds (meeting SLA) Workflow Efficiency High cognitive load, multiple context switches ~25% faster per case, auto-populated structured reports Reporting Accuracy Frequent transcription errors, hallucinations ~12% fewer errors, reliable dictation capture Scale & Concurrency Limited by infrastructure Supports 1,000+ simultaneous microphones, 100-200 reports/day per physician, sub-second STT latency, multi-model orchestration scalable to billions of tokens/month Audit & Compliance Fragmented reporting, siloed data Structured, auditable workflows; RDS/AEP enable interoperability AI Model Performance Generic transcription, reasoning models Supports RADPAIR’s fine-tuned STT model, hosted and orchestrated in real-time via Fireworks infrastructure. Trained on radiology audio, ensures accurate capture of specialized terminology; downstream reasoning and multi-step orchestration are handled by RADPAIR.

RADPAIR’s Mission: Smarter, Safer Radiology Workflows

Radiology departments are critical to modern healthcare, yet reporting workflows remain slow, fragmented, and cognitively taxing. Radiologists switch between multiple screens, dictate findings line by line, and manually retrieve prior measurements or reports. Studies indicate that radiologists spend approximately 36% - 54% of their time on image interpretation, with the rest on administrative tasks ( JACR, 2023 ). This imbalance not only reduces productivity, but also contributes to burnout and can affect patient outcomes. Founded by practicing radiologist Avez Rizvi , RADPAIR aims to redesign radiology workflows. Today, RADPAIR represents 13-14% of the U.S. radiology reporting market , including its largest client, Radiology Partners , supporting 4,000 physicians across 40-50 million cases annually . Their mission is to enable next-generation, agentic AI workflows that orchestrate multiple AI agents from dictation to reasoning to system control so radiologists can prioritize patient care and high-value interpretations.

The Challenge: Fragmented Workflows and Antiquated Standards

Traditional radiology workflows are fragmented and cumbersome: Imaging → Switch screens → Report software → Dictate findings line by line → Manually retrieve prior reports → Compile measurements → Submit report

Key pain points : • Cognitive overload: Multiple context switches slow work and increase fatigue • Siloed data streams: Difficult to incorporate prior studies, measurements, or external information, increasing risk of errors • Outdated standards: Legacy HL7 limits integration with modern AI • AI adoption challenges: Existing tools cannot reliably orchestrate tasks across systems • Administrative burden: Limits focus on patient care

STT-Specific Bottlenecks and Multi-Agent Requirements Scaling AI-assisted transcription is particularly challenging, with constraints that impact physician adoption and workflow trust: • Speed & Perceived Responsiveness : Early cloud-based STT systems were slower and less accurate than on-premise tools like legacy systems, frustrating radiologists. • Domain-Specific Terminology : Radiologists often use shorthand, incomplete phrases, or specialized terms (“AAA” for abdominal aortic aneurysm, “tib-fib” for tibia/fibula). Off-the-shelf transcription models often struggle to capture these nuances accurately, which undermines both trust and clinical precision. At the same time, overly specialized speech engines tuned only for medical or radiology lexicons limit natural, conversational input—preventing users from fully unlocking agentic AI workflows. • Multi-Agent Orchestration : Real workflows require simultaneous execution of multiple models - transcription, reasoning, report...