OctaPulse Uses Computer Vision to Inspect Fish Faster Than Any Human

The Macro: Aquaculture Is Booming and the QA Is Still Manual

Aquaculture is a $300 billion global industry that has been growing faster than wild fisheries for two decades. Fish farms now produce more fish than wild catch. The US alone is pushing to expand domestic aquaculture production to reduce reliance on imported seafood. But the quality assurance methods in fish farming look like they belong in a different century.

At the hatchery level, fish need to be inspected for deformities, disease markers, and growth metrics before they advance to the next stage of production. This inspection happens by hand. A technician picks up each fish, examines it, and makes a pass/fail decision. Five minutes per fish. When you have a hatchery producing tens of thousands of juveniles, the math gets ugly fast. The labor cost is enormous, the throughput is limited, and human consistency degrades over a long shift.

Broodstock phenotyping is even more critical. Selecting the right breeding stock determines the genetics of every fish that follows. Bad selections cascade through generations, reducing growth rates, increasing deformity rates, and lowering overall production quality.

The technology to automate this exists. Computer vision can detect deformities, measure morphological features, and classify fish faster and more consistently than humans. The challenge has been building systems that work in the wet, slippery, high-throughput environment of a working hatchery.

The Micro: A CMU Roboticist and an Aquaculture Industry Leader

Paul Grech and Rohan Singh founded OctaPulse. Paul leads commercial and partnerships and is recognized as a Future Leader in the National Fisheries Institute and Coalition for Sustainable Aquaculture. Rohan leads engineering with a background in robotics and AI from CMU, ASML, Toyota, Tesla, and NVIDIA. They are a two-person team from YC Winter 2026 with Jon Xu.

The product uses AI vision to automate hatchery QA, starting with broodstock phenotyping and juvenile deformity inspection. It cuts inspection time from about 5 minutes to under 30 seconds per fish, with more than 90% accuracy. The goal is to bring automation across the entire fish production lifecycle.

The traction is strong. OctaPulse has signed a six-figure paid pilot with the largest trout producer in the United States, and their model accuracy exceeds 95%. A six-figure pilot from a single customer at this stage is exceptional for a hardware-adjacent startup.

The Verdict

OctaPulse is a perfect example of AI applied to a specific, measurable problem in an underserved industry. The aquaculture market is large, growing, and badly in need of automation. The founding team combines deep industry knowledge with serious engineering capability. The paid pilot proves that fish producers will pay for this.

The risk is hardware scaling. Building robust computer vision systems that work in hatchery environments requires rugged hardware, reliable image capture in wet conditions, and integration with existing hatchery workflows. Each deployment is more complex than a pure software install.

In 30 days, I want to see results from the trout producer pilot. In 60 days, the question is whether other major producers are in the pipeline. In 90 days, I want to know about species coverage. If OctaPulse works for trout, salmon, tilapia, and shrimp, the addressable market expands dramatically.