What They Build

Milliray makes hardware and software for detecting, tracking, and classifying small unmanned aerial systems (UAS). The core product is a millimeter-wave radar sensor designed from first principles to detect low-signature targets, specifically the kind of small, fast, low-altitude drones that conventional radar systems see as noise.

The product stack has three layers:

1. Sensor hardware. Millimeter-wave radar arrays (77-79GHz range) purpose-built for the specific radar cross-section profiles of nano and micro UAS. These are not off-the-shelf automotive radar modules bolted onto a pan-tilt unit. The antenna geometry, pulse timing, and signal processing chain are all tuned for this specific detection problem.
2. Sensor fusion. The radar integrates with camera arrays and acoustic sensors in a distributed mesh. Each sensor type fills in the others' blind spots. Radar provides all-weather range and velocity. Cameras provide visual confirmation and classification. Acoustics provide low-cost perimeter tripwires. The system produces a unified airspace picture rather than three siloed alarms.
3. Classification ML. The real differentiator isn't detection, it's classification. A bird flock, a plastic bag, and a DJI Mini 4 Pro have different radar signatures. Milliray runs convolutional neural networks on the radar doppler return data to distinguish between them. This is what kills false positive rates. Fewer false alarms means fewer operator desensitizations. Fewer desensitizations means humans actually respond when the real alert fires.

The system can be deployed as a standalone unit for a single facility or networked across a site into a distributed detection grid. Scalability is a core design constraint, an airport has very different topology needs than a power station or a forward military operating base.

The Founders

The team is credible in a way that most YC hardware startups are not. Thomas Doherty (CTO) holds a PhD from Oxford and spent a decade working in optical and laser technologies, with a Royal Academy of Engineering Enterprise Fellowship. Jordina Frances de Mas (COO) has a PhD from St Andrews focused on automated reasoning and ML. Matthew Moore rounds out the founding team on the commercial and go-to-market side.

This is a PhD-heavy founding team working on a PhD-hard problem. That's appropriate. The physics of millimeter-wave radar propagation, the signal processing pipeline to extract micro-Doppler signatures from small targets, the architecture of a sensor fusion system that operates in real-time, none of this is tutorial content. It takes years of domain knowledge to do it right.

Why Millimeter-Wave?

Millimeter-wave radar (mmWave, 30-300GHz) sits in an interesting band. Wavelengths are short enough to resolve small targets with compact apertures. Frequencies are high enough to extract meaningful Doppler returns from slow-moving objects like drone rotors. The technology is mature enough to manufacture economically, automotive mmWave radar ICs from Texas Instruments and Infineon have driven costs down by orders of magnitude over the past decade.

The critical physics insight is micro-Doppler. When a drone's rotors spin, they create a characteristic Doppler frequency spread around the main body return. This signature is distinct. It looks different from a bird's wing beat, different from a car's rotating tire, different from a helicopter's larger blade geometry. If you train an ML model on enough labeled radar captures of different UAS types, you can classify targets with high accuracy in real-time.

The challenge is environmental clutter. Urban environments are radar nightmares. Ground clutter from buildings, moving vehicles, trees in wind, all of this creates noise that overwhelms naive detection algorithms. Getting robust performance in real-world deployments, not just controlled test ranges, is where most academic mmWave drone detection research falls apart. It's also where Milliray is focused.

The Market

The counter-UAS market is valued at roughly $10 billion in 2026 and projected to reach $70 billion by 2034. That growth rate, 26% CAGR, is not driven by optimistic analyst assumptions. It's driven by actual events: drone attacks on Ukrainian power infrastructure, drone incursions over French nuclear plants, drone disruptions at UK airports, and the proliferation of commercial off-the-shelf drone technology that is outpacing defensive capabilities.

Customer segments break down clearly:

• Airports and aviation. Regulatory pressure after Gatwick. ICAO standards under development. High willingness to pay. Long procurement cycles.
• Critical infrastructure. Power stations, water treatment, data centers, refineries. Nation-state and domestic terrorism threat vectors. Typically managed by facility security teams with dedicated budgets.
• Defense and government. Military forward operating bases, border patrol, event security. Highest unit values. Slowest sales cycles. Requires relevant certifications and often domestic manufacturing.
• Venues and VIP protection. Stadiums, concerts, heads of state. Faster procurement, lower unit values, volume opportunity.

The go-to-market path for a hardware startup in this space is genuinely hard. Defense procurement can take three to five years from first contact to deployed contract. Airports are similarly slow. The practical path is to land venue and VIP protection deals first, faster sales cycles, real deployments, real data, real case studies, then use those references to unlock the defense and aviation channels.

Difficulty Score

Overall difficulty: 8/10. This is a hard problem. Not "hard" in the sense of requiring clever architecture choices on AWS. Hard in the sense that the core technology requires domain expertise in radar physics, signal processing, and embedded ML that you cannot acquire by reading documentation.

The Moat

What's genuinely hard to replicate:

• The dataset. Every deployment generates labeled radar captures of real drone activity in real environments. This data is rare, valuable, and not available publicly. The more deployments Milliray has, the better their models get. This is a data flywheel that compounds over time.
• The physics expertise. Thomas Doherty's decade in optical and laser technologies and Jordina's ML expertise aren't replaceable with a recruiting budget. The specific combination of radar hardware design, signal processing, and applied ML is a genuinely scarce skillset.
• Customer relationships in regulated markets. Selling to airports and defense requires trust, clearances, and references. Once you're in, you're sticky. Switching costs are high when the alternative is re-certifying a new vendor through a multi-year procurement process.

What's replicable:

• The hardware stack. mmWave radar ICs are commodity components. A well-funded competitor can buy the same chips and build a competing sensor.
• The general software architecture. Sensor fusion with camera and acoustic integration is a known pattern.
• The ML approach. Micro-Doppler classification with CNNs is published in academic literature. A team with the right skills could implement it.

The honest assessment is that this isn't a winner-take-all market. Airports in the UK might buy from Milliray. Airports in France might buy from a French defense contractor. The US government almost certainly requires domestic manufacturing. Milliray's strategic positioning matters enormously, geographic focus, certification strategy, and which vertical they dominate first will determine whether they're a standalone defense tech company or an acquisition target for a Thales or L3Harris.

The Bottom Line

Milliray is the kind of startup that doesn't get the flashy TechCrunch profile because their product doesn't have a consumer interface. There's no iOS app, no freemium tier, no viral moment. It's a box you bolt to a fence post at Heathrow that quietly keeps the airspace clean.

But the drone threat is real, it's accelerating, and the incumbent solutions are genuinely inadequate. The founding team has the domain expertise to build something that works in real conditions, not just on test ranges. And they're building in a market where willingness to pay is high and switching costs are sticky.

If they can crack the defense procurement cycle and get into the data flywheel early enough, this could be a serious company. The technology is genuinely hard. The market is genuinely large. And the problem is not going away, if anything, the proliferation of cheap drones makes it worse every year.

This is exactly the kind of company YC should be funding.