An innovator in healthcare AI is working to deploy autonomous medical agents to streamline medical note generation. These systems promise to reduce clinician workload, improve documentation accuracy, and accelerate patient care. But to make them effective in the real world, the models behind them need structured, expert-reviewed, multilingual data pipelines.

That’s where Perle comes in.

‍

The Challenge

Medical note generation is one of the most promising applications of AI in healthcare, but it also faces steep challenges:

• Complex data flow: Multi-agent RAG (retrieval-augmented generation) systems depend on accurate context and seamless coordination. Errors can cascade across agents, lowering output quality.
  
  
• Edge cases in medicine: Generic benchmarks do not reflect the nuances of real-world medical scenarios, which risks reducing trust among clinicians.
  
  
• Multilingual needs: Healthcare is multilingual by nature. Notes and patient interactions in Arabic, Spanish, and other languages require equally rigorous review and quality assurance.
  
  

This medical AI company needed a partner who could bring structure, rigor, and domain expertise to the data powering their autonomous medical agents. The company also needed to ensure quality — which is why they came to Perle. 

‍

The Solution

Perle partnered with the company across three critical areas:

1. Optimizing the RAG-Based Multi-Agent Architecture
   Perle supported improvements to the generative stack by tightening the quality of context retrieval and improving coordination across agents, ensuring medical notes were accurate and contextually aligned.
   
   
2. Human-in-the-Loop Benchmarking
   Working with Perle’s diverse network of medical experts, the medical AI company co-developed a benchmarking framework grounded in real-world edge cases. This approach provided reliable guardrails for evaluating AI outputs where precision matters most.
   
   
3. Multilingual Review and QA
   Perle provided expert annotation, review, and QA in multiple languages—including Arabic and Spanish—to ensure the company’s  medical agents could serve diverse patient populations without compromising quality.

‍

The Results

Perle only uses domain experts — not general annotators. Perle sourced doctors that have already worked on AI projects, so they understood the medical side, and the AI side of the project.

“Our doctors know what customers are trying to solve for,” said Moe Abdelfattah, Head of Product Operations at Perle. “Our doctors bring wisdom and perspective.” 

By integrating Perle’s expertise into their pipeline, the medical AI company achieved:

• Improved accuracy and reliability of generated medical notes, even in complex multi-agent workflows.
  
  
• Robust benchmarking frameworks validated by medical experts, helping Sully test and refine their systems in real-world contexts.
  
  
• Multilingual readiness, expanding the accessibility of Sully’s medical agents to global markets and diverse patient populations.
  
  

With Perle’s high-quality annotation and expert-in-the-loop workflows, the company can now build AI that is not only more accurate but also more trusted, scalable, and inclusive.

Get in touch to start your next project.