10x'ing one of the hardest problems in clinical AI. Why we doubled down on CARPL.

Where clinical AI actually breaks

One of our long-held beliefs in healthcare AI is that the hardest problem is rarely the algorithm itself. It is everything around it that determines whether the algorithm ever gets used.

Take a hospital looking to adopt AI for radiology. There are over 1,100 FDA-approved AI algorithms for radiology, with more entering the market every month. Different specialists want to be able to evaluate different algorithms, compare performance, and then choose what works best for their practice. They work only inside their PACS so every output has to also land inside the PACS for the algorithm to be usable. 

To adopt ten AI algorithms, a hospital has to run ten integrations, ten evaluations, ten procurement cycles.

Building the missing layer 

That is precisely the problem CARPL set out to solve. CARPL’s platform sits between the PACS and AI vendors, giving hospitals a single platform to evaluate, procure, deploy, and monitor radiology AI. Instead of integrating separately with every vendor, the hospital integrates once with CARPL and gains access to a growing ecosystem of AI algorithms.

Our original investment thesis was that the bottleneck to adoption of Clinical AI is not model quality but integration of those algorithms into complex hospital workflows. And as the number of AI algorithms explode, the value of the aggregation layer would only increase.

When the thesis meets reality

Two years later, that thesis has largely played out.

The number of FDA-approved imaging algorithms has grown ~5x (221 in 2023 to 1,163 in 2026). The combinatorial complexity we anticipated has arrived. 

Over the same period, CARPL has grown to serve some of the world's leading healthcare institutions, including Radnet (USA), I-Med (Australia), Fleury (Brazil), Singapore Government, MD Anderson (USA). Today, their platform hosts 300+ algorithms and processes more than 130,000 patient scans every month.

Over the past nine months, the average number of algorithms deployed per hospital has roughly doubled. Once a hospital is integrated into the platform, adopting the next AI application becomes dramatically easier. That is exactly the flywheel we hoped would emerge.

Strategic decisions that changed the game

Several of the team's early strategic decisions deserve recognition because none of them were obvious at the time.

Partnering with the gatekeepers

Channel partnerships are notoriously difficult for early-stage companies. But CARPL recognized that PACS providers were the biggest gatekeepers to AI adoption. Instead of competing with them, the team aligned incentives. Today, CARPL partners with all 10 of the world's largest PACS providers, and many others besides. The biggest barrier to adoption has become one of its strongest distribution advantages.

Winning the toughest customers first

Rather than building credibility through smaller customers, CARPL went directly after the largest and most prestigious hospital systems from day one, despite them being the hardest accounts to win. The sales cycles were longer, but the strategic value of those reference customers was enormous. Today, three of the world's largest radiology service providers are customers.

Going global before going local

The third was taking a global-first approach. Instead of focusing on the US immediately, CARPL prioritized markets such as Australia, Singapore, and the UK, where hospitals were adopting AI more rapidly and regulatory friction was lower. The team also chose not to make India its primary market, despite securing a significant early contract there. That discipline about where to plant the flag first is rarer than it sounds.

Raising the talent bar

The company also strengthened itself with every financing round, bringing in experienced leaders across Product, Product Marketing, Sales, Engineering, and Medical Affairs. One of the strongest signals we look for in founders is their ability to consistently attract exceptional talent. CARPL has done exactly that.

Backing a category-defining founder

Behind every one of these decisions is the same person: Vidur.

Founders with Vidur's combination of domain expertise, technical depth, and an intense desire to change the world are exceptionally rare.

He is a physician by training. He helped build one of India's leading diagnostic businesses. Alongside some of the world's leading researchers, he has published research on radiology AI. He also has an MBA from Wharton.

But what stands out isn't the résumé. It is his ability to move seamlessly between deeply technical discussions and nuanced clinical conversations. Very few founders can go from Kubernetes to lung cancer staging in the same conversation. Building this category requires exactly that combination.

Seed to Series A: Conviction strengthened by execution

We invested in CARPL at Seed because we believed the adoption bottleneck was integration, not intelligence. We invested again at Series A because their execution has only strengthened that thesis. 

In a recent conversation, Vidur told us that the average radiologist examines around 100 scans a day. His ambition is to make that 1000 over the next five years. If he cannot do that, he feels he would have failed. That relentless drive to redefine what is possible in clinical AI is what excites us.

‍

Read the media coverage here.

‍