GLOBO Case Study: AI Quality Monitoring for Medical Interpretation

About GLOBO

GLOBO Language Solutions, the leading B2B provider of interpreting and translation services, provides real-time interpretation for healthcare organizations, connecting LEP (Limited English Proficiency) patients with their network of 10,000+ linguists who speak more than 430 languages and dialects via video calls. Their HQ platform manages thousands of interpretation sessions every day across hospitals, clinics, and telehealth providers.

When CEO Dipak Patel set out to bring AI into the business, he was looking for a long-term strategic partner. “I just didn’t want to outsource to a bunch of engineers,” he says. He wanted three things: real expertise, a real partner, and dependability.

The challenge

Quality assurance that couldn’t keep up.

Before this engagement, GLOBO had no automated way to monitor interpretation quality in real time. Whether interpreters were presenting professionally on camera (proper lighting, framing, a clean background, a professional setting) was reviewed manually, after the fact, on a small sample of sessions.

Quality assurance was manual, sample-based, and retrospective. A process that’s difficult to scale, and one that left compliance gaps in a sector where miscommunication carries real patient-safety risk.

Why Progression Labs

Three things, in Dipak’s words.

Dipak is a former engineer and former consultant: someone who values an outside-in perspective when an internal team risks getting too close to the work. GLOBO had a strong engineering team, but the company saw the opportunity and wanted to optimize for speed, and that’s why they looked for an external partner to move faster without hiring a department to keep up. He set three bars for any partner.

Experts in an evolving field

“I needed people that were experts in this evolving technology. It was moving so quickly. People aware of the latest trends with a perspective on where it would go.”

A partner, not a vendor

“Someone who would not just say ‘yes, yes, yes,’ but help challenge us, take time to understand where we’re going as a business, and help us in different situations.”

Dependable

“In order for us to be successful, we have to move really quickly. I need a partner I can call up and say, hey, I need this, and know it’s going to get done.”

“They’ve gone beyond just a number of engineers that I’m outsourcing AI for. They truly have been a thought partner. Someone dependable, someone who has actually helped us shape our strategy at GLOBO.”

Dipak Patel · CEO, GLOBO

The work

Using Claude’s multimodal capabilities to solve complex challenges.

Responding to the scaling challenges, GLOBO asked the question the industry had written off: could AI watch every interpretation session as it happened, and flag quality problems in real time? Within a short timeline we had a working prototype, and it turned into Live Quality: an event-driven microservices platform on AWS ECS Fargate, built on the internal GAIT toolkit.

In the backend we used Claude on AWS Bedrock. Image analysis evaluates interpreter screenshots against a structured rubric for lighting, framing, background, and professionalism. The result is a system that monitors for visual quality on every call, instead of the manual sample-based review that came before. Shipped in production with 1,500+ unit tests.

Three reasons we picked Claude for GLOBO.

Before we committed, we evaluated Claude head-to-head against other leading models on the production task. Claude won on three fronts that mattered in a regulated healthcare environment.

Structured-output reliability

Every Claude response feeds straight into a webhook payload that GLOBO’s HQ platform parses programmatically. Claude’s instruction-following for JSON schemas was measurably more consistent than the alternatives: fewer schema violations, fewer hallucinated fields.

Vision for the screenshot rubric

Claude evaluates interpreter screenshots against a structured rubric (lighting, framing, background, professionalism) in a single multimodal prompt. Competing models needed heavier preprocessing or produced less granular assessments.

AWS Bedrock: data stays in GLOBO’s VPC

GLOBO runs entirely on AWS. Bedrock keeps every Claude call inside their cloud boundary; no patient-adjacent data ever leaves their VPC. Cross-region inference profiles give resilience without compromising the data perimeter.

The outcome

Every session monitored.

QA monitoring went from a single-digit manual sample to 100% automated coverage across every interpretation session. Live Quality is in production, watching every call as it happens.

Looking for the same kind of partner?

Engagements begin with the capability assessment. It builds an evidence-based understanding of where you are in your transformation, and of what you need to do to achieve your goals.

Start a conversation