Kaliber Labs

Bottling What the Best Surgeons Know

Here is a fact about surgery that is both obvious and slightly absurd: some of the most valuable information in medicine is generated inside a live operating room, streamed from a camera on the end of an arthroscope, and then almost entirely thrown away. The surgeon watches the feed, does the work, and later types up a note from memory. The video - the actual record of what happened, frame by frame - typically goes nowhere. Kaliber Labs looked at that pile of discarded footage and saw, reasonably, a business.

Kaliber Labs is a San Francisco company founded around 2018 by Ray Rahman, and its self-description is refreshingly free of the usual healthcare-AI fog. Its tagline is “We manufacture expertise.” That is a strange sentence if you sit with it. Expertise is supposed to be the thing you cannot manufacture - the twenty-years-of-cases judgment that lets a good arthroscopic surgeon glance at a smeared, bloody, magnified view of a shoulder joint and instantly know what they are looking at. Kaliber's wager is that this judgment, or a useful slice of it, can be captured from enough examples and re-emitted as software.

To do that, you need the examples. This is the part competitors cannot simply copy over a weekend, and it is the part Kaliber has spent years assembling: one of the world's largest libraries of arthroscopic surgical video, annotated and guided by more than 45 of the top orthopedic surgeons in the United States. That surgeon roster is not decoration. In a domain where a wrong label is not an embarrassing chatbot answer but a clinical claim, the people who train and validate the model matter as much as the model itself. Kaliber's proprietary system learns to recognize anatomy, measure structures, and place landmarks in real time - the visual grammar that a surgeon internalizes over a career, rendered as something a machine can perform on demand.

Two jobs, one camera

What the company actually sells splits neatly into two halves, and the split is instructive. The first half faces the operating room. Kaliber's surgical guidance system - it has gone by the name DSS, for digital surgical solutions - provides real-time perception during a procedure: it watches the arthroscopic feed, recognizes what is on screen, and offers case awareness the way a very attentive second set of eyes might. There is a feature the company calls 3D surgical GPS, which lets a surgeon visualize complex anatomy in three dimensions while operating. The pitch is not “the AI does the surgery.” The pitch is “the AI understands the video, so the surgeon has more to work with.”

The second half faces the back office, which is where a surprising amount of the value hides. After surgery, someone has to write the operative note - a detailed, structured account of what was done, which drives billing, records, and continuity of care. It is time-consuming, it is done from memory, and it is exactly the kind of task a system that already watched the whole procedure should be able to draft. Kaliber's Kapitan-powered AI agents generate the OpNote, orchestrate surgical workflow, and handle a set of chores for surgeons, surgical technicians, and administrators. There is also a patient-facing thread: personalized communication with AI-labeled intraoperative imaging, so the person who was unconscious for the whole event can actually see and understand what happened inside their own knee or shoulder.

A foundry, not a product

The most interesting thing about Kaliber in its current form is that it has stopped describing itself as a single product and started describing itself as a foundry - an AI research shop that develops systems for specialized domains and then spins them into focused companies. The stated research areas read like a graduate seminar: computer vision, clinical science, cognitive science, robotics, world models, neuro-symbolic reasoning. The output is a small constellation of ventures that share one core idea. DSS handles surgical guidance. Veris handles the instrument lifecycle, watching the sterile processing department and accounting for every instrument used inside the OR - an unglamorous, high-stakes counting problem that hospitals genuinely lose sleep over. Citrus is socially intelligent AI for clinical and commercial conversations. Movendi builds physical, embodied AI models that actuate clinical robotics across care environments.

You can be skeptical of the foundry framing - it is, after all, a nice way to describe a company that has several products - but the underlying logic is coherent. Each venture is the same move applied to a different surface: take a form of expert judgment, learn it from data, and embed it in a system that perceives, reasons, and acts. If you believe that domain-specific AI beats general-purpose AI in high-stakes settings, then owning the expensive-to-acquire data and the specialist relationships in one vertical, and reusing the platform across adjacent problems, is a defensible way to build.

The founder detour

Ray Rahman's path to surgical AI did not run through medicine. He holds degrees in mechanical engineering and economics from the University of Rochester and an MBA from MIT Sloan, and his earlier career was in private equity, fintech, and big data, with a couple of acquisitions along the way. The line he is proudest of is genuinely unusual for a healthcare-software founder: he structured a new asset class, micro-credit-backed securities, and the BRAC MCBS Series 1 earned a AAA rating while channeling entrepreneurial capital to more than 1.2 million poor households. The through-line from AAA-rated microcredit to arthroscopic AI is not the industry. It is the instinct to take something rare, illiquid, and locked inside a few institutions, and turn it into something that scales.

On the money: figures vary by database, which is normal for a private company. PitchBook puts total funding at roughly $21.7 million, with a Series A tranche of about $3.5 million landing in August 2022. The team is around 64 people, works out of San Francisco, and supports remote roles. None of these numbers scream unicorn, and that is arguably the point. Software-as-a-medical-device is a domain where moving fast and breaking things is a malpractice suit, and where trust is accumulated slowly, surgeon by surgeon. Building here is supposed to be deliberate.

Who it is for, and what they get

The people on the receiving end of all this are not, mostly, consumers. Kaliber is a business-to-business company, and its users cluster into four groups: the surgeon, the surgical technician, the surgery-center administrator, and - one step removed - the patient. A surgeon gets a system that recognizes anatomy and drafts the operative note. A tech gets an instrument record that reconciles itself. An administrator gets workflow orchestration and a cleaner back office. A patient gets a version of their own procedure they can actually look at and understand, with the intraoperative images labeled by AI rather than left as an inscrutable grid of pink. If you have ever left a medical appointment with a bill and a shrug, that last one is quietly radical: it treats the patient as someone entitled to know what happened.

The competitive neighborhood is real and getting crowded. Surgical-video AI and computer-vision-for-the-OR companies - names like Theator, Caresyntax, Proprio, and Activ Surgical - are all circling adjacent problems, and the larger surgical-robotics players loom over the whole category. Kaliber's answer to the crowding is the same as its answer to everything else: go narrow, go deep, and own the data. Arthroscopy is a specific, high-volume, camera-native kind of surgery, which makes it an unusually good beachhead for a company that believes its advantage compounds with every additional hour of annotated footage and every additional surgeon who signs on.

Which is the quietly compelling thing about Kaliber Labs. It is not promising to replace surgeons, and it is not promising a magic general model that solves medicine. It is doing something narrower and more grounded: watching the video that everyone else deletes, learning the visual language of a specific kind of surgery from the people who are best at it, and handing back both real-time perception and the paperwork nobody wants to do. Whether “manufacturing expertise” turns out to be a slogan or a product category, the raw material is real, the surgeons are real, and the footage - finally - is being kept.