He didn't build an AI to read your scan. He built one to kill the seven hours of busywork standing between the radiologist and the read.
Most founders pitch you a future. Shiva Suri starts with a basement. Early in the pandemic he was working from home next to his mother, a world-class radiologist, and he watched her read a single case by hopping between five different PACS platforms, repeating the same clerical ritual case after case. He timed it. Seven, sometimes eight hours of her day went to administrative friction. The actual radiology - the part she trained decades for - got maybe five percent of the clock.
That ratio is the whole company. New Lantern, which Suri founded in 2021 and runs as CEO, is built on a contrarian read of the AI-in-medicine gold rush: everyone else wants to point a model at the image and play doctor. Suri pointed his at everything around the image instead. The dictation. The report drafting. The measurements. The worklist. The metadata that a human keeps re-typing off a technologist's worksheet. Automate that, and the radiologist gets to stay a radiologist.
The platform is named Curie. It folds three traditionally separate products into one screen - a smart worklist, a cloud-based PACS viewer, and AI-powered report generation - and the company says it lets a radiologist clear twice as many cases in the same hours. New Lantern calls the result an "AI radiology resident": the eager junior who does the legwork, drafts the first pass, and hands the senior physician something to sharpen rather than something to build from scratch.
Suri is an engineer by training and temperament. He earned a bachelor's in computer engineering and a master's in computer science from the University of Pennsylvania, then went to Confluent - the company commercializing Apache Kafka - where he worked on cloud-native data infrastructure. That is an unusual on-ramp to healthcare. It is also exactly the right one. Radiology's bottleneck is not a lack of clever image models; it is a tangle of legacy software that was stitched together over twenty-five years. Suri treats it as a systems problem, because that is what it is.
His framing is blunt. "A radiologist's job was supposed to be playing Sherlock Holmes in images," he has said, "not constantly mouse-clicking all over their PACS." He is not chasing an incremental plug-in. He wants New Lantern to be the biggest evolution in the radiology software market since PACS itself arrived a quarter-century ago - and he is willing to rebuild the whole stack to get there rather than bolt features onto someone else's.
The pitch would be easy to dismiss as founder theater if the numbers weren't starting to show up. At SIIM 2025, the radiology informatics conference, New Lantern presented a real-world case study with results that read like a stress test of its own thesis: dictated words dropped 72% (from 205 to 57), navigation actions fell 97% (from 32 down to a single click), and radiologists recovered roughly 100 minutes per shift. The platform automates over 75% of the non-diagnostic work - the stuff nobody went to medical school to do.
Investors noticed. In November 2024, New Lantern closed a $19 million Series A led by Benchmark, with general partner Eric Vishria taking a board seat - part of $22.6 million raised in total. The cap table reads like a who's-who of builders rather than passive money: Vercel's Guillermo Rauch, Replit's Amjad Masad, Confluent co-founder Jay Kreps, prolific angel Gokul Rajaram, Benchling's Saji Wickramasekara, plus Afore Capital, Anthology Fund, Neo, and SV Angel. When the people who built the developer tools you use every day write checks into your radiology startup, it is usually because the engineering smells right.
The phrase does a lot of quiet work. A resident is not a replacement for the attending physician; a resident is the person who pre-reads the case, lines up the priors, drafts the report, and flags what matters so the senior doctor can spend their attention where it counts. That is the relationship New Lantern is trying to manufacture in software. Automatic dictation. AI-generated report impressions. Smart segmentation. Adaptive viewing protocols that set up the study the way the radiologist would have, before they ask. Metadata pulled straight off the technologist's worksheet so nobody re-types it. The radiologist still makes every call. They just stop doing the resident's chores themselves.
It is worth sitting with how counterintuitive that is in 2025. The loud version of medical AI promises a model that out-diagnoses the human. Suri's quieter version assumes the human is the point and engineers everything else out of their way. The two systems New Lantern fuses - PACS, the picture-archiving layer where images live, and the reporting software where findings get written - have been bolted together awkwardly for a generation. Collapsing them into one screen is less glamorous than a diagnostic breakthrough and arguably more useful, because it attacks the friction that every radiologist feels every single day.
The company has not stood still since the raise. In August 2025 New Lantern widened the platform beyond general imaging with dedicated mammography and PET/CT viewers - specialized reading modes for breast imaging and nuclear medicine, two of the more demanding corners of the field. The move signals the ambition under the "automate 90%" slogan: not a single clever feature, but an all-in-one operating layer for how radiology actually gets read and reported. Specialized viewers are a tell. They mean the company is going deep into the workflows that resist generic tooling, not just skimming the easy cases.
There is a reason the investor list reads like an engineering roll call rather than a healthcare one. Builders who have shipped infrastructure - the people behind Vercel, Replit, Confluent's Kafka - tend to recognize a fellow systems thinker, and they tend to bet on execution over narrative. New Lantern's twenty-nine-odd person team, described by the company as Silicon Valley engineers, AI researchers, and radiologists working together, is built to that taste: people who can both write the code and sit with the doctor whose day it changes.
What makes Suri worth watching is the discipline of the wager. It would have been easier, and far more fashionable, to build a diagnostic model and ride the AI headlines. He chose the unglamorous middle - the clicks, the dictation, the metadata - because that is where the radiologist's day actually leaks. The bet is that the future of medical AI isn't the machine that replaces the expert. It's the one that finally lets the expert do the thing only they can do. He learned that watching his mother. Now he is trying to give it back to every radiologist who reads a scan.
New Lantern put a single real-world read under the microscope. Here is how much friction the platform stripped out. Bigger bar = bigger reduction.
A radiologist's job was supposed to be playing Sherlock Holmes in images, not constantly mouse-clicking all over their PACS.
We want New Lantern to be the biggest evolution in the market since PACS was introduced 25 years ago.
The muse was his mother. New Lantern exists because a radiologist's bad workday had a software engineer for a son.
The platform is called Curie - a nod to Marie Curie, whose work helped make medical imaging possible in the first place.
He came from Confluent, the data-streaming company behind Apache Kafka. Healthcare wasn't the obvious next move. It was the right one.
He pointedly does not use AI to read images. He uses it to delete the admin around the read - the opposite of the headline play.
His angels build the tools you use: Vercel's Guillermo Rauch, Replit's Amjad Masad, and Confluent's own Jay Kreps.
Same hours, double the cases. That's the bet New Lantern makes to every radiology practice that signs on.