Mosaic raises $47M to build machine-learning surgical navigation

By James Calloway

What did Mosaic announce?

Mosaic Intelligence Labs emerged from stealth on November 10, 2026 with $47 million in seed funding (early-stage capital) to develop sensing-enabled spatial intelligence (a live representation of anatomy and instruments) and robotic systems for surgery. The San Francisco startup will initially focus on endovascular procedures, operations conducted through blood vessels.

Its first target is also its data strategy: recording the environment, instrument movement and physician actions for future machine learning (systems that improve from data). The round names 8VC, Fundomo, Lingotto Innovation and Obvious Ventures among its backers.

The announcement does not identify a lead investor, valuation, use of funds or commercialization timetable. Those omissions leave the round’s financial structure and deployment schedule uncertain.

Chief executive Ileana Pirozzi and chief technology officer Alison “Ali” Kight Mackanic founded Mosaic after meeting during doctoral studies at Stanford. According to Mosaic, their work alongside clinicians showed that less-invasive procedures still depend on knowledge held by individual operators.

Pirozzi said: “We kept returning to one question: what would it take to scale the knowledge of the very best surgeons?” She added: “We believe that starts with making our machines aware, and turning surgical repetition into experience we can learn from.”

That promise puts the burden on navigation before it puts the burden on autonomous action. Mosaic says the nearer objective is to assist a physician’s perception and control, not replace the operator. For now, Mosaic has announced a funded development plan, not a clinical result.

Why start with blood-vessel surgery?

Endovascular care lets physicians thread flexible wires and catheters through blood vessels to reach treatment sites. Operators have no direct line of sight inside those vessels, so they work from live X-ray imaging, preoperative scans and the feel of their instruments.

Mosaic identifies radiation exposure and outcomes that vary with operator experience as limitations of that workflow. The company says interpreting those signals takes years of practice, and much of the resulting knowledge resists recording or transfer.

Access has therefore become less invasive while navigation still depends heavily on the person holding the instrument. That mismatch creates the opening Mosaic wants to address, although the announcement does not quantify how large the performance gap is.

Mosaic’s website compares three access routes:

• Open surgery uses a 20–30 cm incision.
• Keyhole surgery uses 1–2 cm ports.
• Endovascular surgery uses a 2 mm needle puncture.

Those figures compare access size; they do not establish lower total risk or better outcomes. Mosaic’s release provides no clinical evidence for a benefit claim either.

Mosaic also lists four current endovascular applications:

• Stopping strokes.
• Clearing blocked coronary arteries.
• Repairing aneurysms.
• Restoring blood flow to limbs.

The company describes blood vessels as routes to virtually every organ. That reach supports Mosaic’s plan to treat vascular navigation as a capability that can carry across several procedures, rather than a tool built for one operation.

In plain terms, Mosaic wants to reduce reliance on individual operator judgment while preserving minimally invasive access. Its systems still need to demonstrate that they can do so.

How would Mosaic’s system learn?

Mosaic intends to instrument the point where surgical tools meet anatomy. Sensor readings would feed computation that reconstructs the vascular environment from within, maintaining a three-dimensional picture of anatomy and instrument position as both change.

A robotic interface would support precise instrument control. The announcement describes this as an intended capability, not a deployed feature, so no claims about accuracy, latency or reliability can yet be assessed.

The proposed learning architecture would record the environment, instrument movement and physician actions as synchronized streams. Physical models would then help the system relate what it senses to what happens when it moves.

Mosaic connects that design to the robotics concept of a world model: a representation of an environment that helps a machine predict how actions will change it. The term describes the proposed learning architecture, not a tested model with published results.

The design connects two problems at once: locating instruments inside anatomy and learning from each recorded procedure. Cumulative learning is the second milestone, not an immediate clinical benefit.

Where does the evidence stop?

The announcement offers an architecture, not clinical validation. Mosaic says all of the sensing, reconstruction, learning and robotic capabilities remain development goals. It supplies no patient-outcome findings or clinical-performance results.

The nearer objective is to help physicians perceive anatomy and control instruments more reliably. Recorded procedures are also intended to become a source of cumulative learning for machines over time.

That sequence matters because dependable navigation must work before recorded procedures can supply useful training data. Mosaic makes that dependency explicit, making cumulative learning a second milestone rather than an immediate benefit.

No measured reduction in radiation exposure appears in the announcement. Nor does Mosaic report comparative performance against experienced operators, despite naming operator variation as a core problem. Those omissions leave its central benefit unquantified.

The $47 million supports an engineering and clinical-development program; it does not establish safer or more effective surgery.

Who is Mosaic hiring?

Mosaic is recruiting engineers, physicians and creatives in San Francisco. The mix reflects a program spanning physical instruments, software, robotics and clinical practice, according to the company.

Its open roles cover hardware engineering, sensing, perception, machine learning and clinical engineering. Hiring across those five areas shows that Mosaic expects hardware, software and clinical work to proceed together.

The announcement names no clinical partner, regulatory milestone or first procedure date. It also gives no prototype maturity or deployment timetable.

The staffing plan signals technical breadth, not deployment readiness.

What could vascular navigation enable next?\n

Mosaic’s longer-term goal is to widen the treatments delivered through vascular routes. The company says catheters can deliver drugs, energy, biologics and other therapies directly to a target site.

More reliable navigation could expand both the range of treatments delivered through blood vessels and the settings in which clinicians can offer them. That would make navigation infrastructure, in Mosaic’s framing, a platform for several therapeutic uses.

The near-term product challenge and longer-term data ambition remain linked. Mosaic says dependable navigation must work in individual procedures before those procedures can supply a useful foundation for cumulative machine learning.

That is a demanding order of operations. Mosaic must first show that its sensors, reconstructed anatomy maps and robotic controls can assist trained clinicians before its recordings can support claims about machine learning at scale.

The company has described a technical route and backed it with $47 million. The next decisive evidence will be clinical-performance and patient-outcome results; without them, Mosaic remains a funded development program rather than a proven surgical system.