Congruent

Congruent was founded in 2025 by a team of silicon-valley veterans specialized in signal processing and machine learning. Clement Barthes, the CEO, previously served as an ML engineer and manager at Zendar and was the CTO of Safehub, where he developed smart sensors to detect structural damage in buildings after earthquakes. His co-founder, Evan Carnahan, is a deep-tech researcher with extensive experience in sensor fusion and neural architectures.

The duo joined the Y Combinator Winter 2026 (W26) batch to scale their vision of AI-first sensing. Based in San Francisco, the founders are leveraging their background at UC Berkeley's PEER lab and their experience in autonomous startups to challenge the dominance of legacy hardware giants. They believe that for AI to truly "act" in the physical world, it must first be allowed to "see" the world in its rawest form.

Congruent is built specifically for Autonomous Vehicle (AV) Developers and Robotics Engineers who are frustrated by the limitations of "off-the-shelf" radar sensors. It is the ideal solution for EV Startups and Tier 1 Suppliers that want to achieve "Level 4" autonomy without relying on expensive, weather-sensitive LiDAR systems.

It is also a critical tool for Industrial Robotics Labs. From autonomous delivery bots to massive agricultural machinery, any robot that operates in unpredictable, outdoor environments can use Congruent to maintain a high "safety buffer." In 2026, it has also become a sanctuary for AI Researchers who need high-fidelity radar simulation data to train the next generation of multimodal foundation models.

Congruent acts as the Data Conduit for autonomous intelligence. Its primary capability is the Raw Data API, which streams unfiltered radar reflections directly into a vehicle's central computer. This allows your neural network to identify subtle environmental cues—like the specific vibration of a cyclist's wheels or the movement of a pedestrian behind a thin fog—that processed point clouds would normally delete.

For the development phase, its Physics-Native Simulator allows for massive-scale parallel testing. You can simulate millions of driving scenarios in "edge-case" weather—such as a blizzard at midnight—and generate radar data that is statistically identical to what the hardware would record in the real world. This "zero-latency" feedback loop allows developers to iterate on their driving policies 10x faster than through road testing alone.

As an early-stage hardware and software company in 2026, Congruent’s pricing is Project-Specific and Scalable. For R&D teams, they offer Development Kits that include a set of prototype radar units and a 12-month license to their simulation environment.

For Automotive OEMs, Congruent provides a high-volume pricing model designed to fit into the bill-of-materials (BOM) for standard consumer vehicles. Their goal is to make AI-capable radar cheaper than high-end cameras, ensuring that every car on the road—from budget commuters to luxury sedans—can be equipped with the same high-fidelity sensing technology.

The most important thing to watch with Congruent is the Onboard Compute Requirements. Because you are processing "raw" data rather than pre-filtered points, your vehicle’s AI chip (NVIDIA Orin, Tesla FSD, etc.) will face a higher data load. Ensure your hardware architecture can handle the increased throughput. Additionally, pay attention to Regulatory Compliance. While radar is a standard automotive sensor, the use of "raw-data" end-to-end models is still being evaluated by safety agencies like the NHTSA, so your legal team should be involved in the integration process.

In the 2026 landscape, the primary alternatives are Arbe and Metawave, both of which offer high-resolution 4D imaging radar. However, these are often more focused on hardware "resolution" rather than the end-to-end "AI training" compatibility that Congruent prioritizes. Zendar (where the founder previously worked) is another competitor, though it focuses more on specific software-defined radar rather than a unified simulator-to-hardware stack. For teams with massive budgets, Luminar or Innoviz LiDAR remain the "gold standard" alternatives for depth, though they are significantly more expensive and less effective in heavy rain.