EN
Back to the archive

The archive · Hardware & Devices · Product decision · 2025–2026

OctaPulse bets computer vision can replace manual fish-farm inspection

YC W26 startup automates hatchery fish QA with CV and robotics; Riverence pilot cut inspection from ~5 min to under 30 s per fish

OctaPulse

The betFarmers will buy CV and robotics for fish inspection once accurate and cheap enough; the measurement data unlocks breeding gains aquaculture never had.Live

What the business is

Builds an AI-vision platform plus soft-gripper robotics that automate hatchery QA for vertically integrated finfish farms: broodstock phenotyping and juvenile deformity inspection, sold as a drop-in service.

How it started

Paul Grech (ex-Bloomberg, CMU MBA) and Rohan Singh (ex-Tesla, Nvidia, ASML) met through CMU ocean-tech events. Grech's research showed a ~$350B aquaculture industry still inspecting by hand: net a few dozen fish, anesthetize them, measure one by one at about 5 minutes each, then extrapolate to populations of hundreds of thousands.

What happened

After YC W26, OctaPulse signed a 6-figure paid pilot with Riverence, North America's largest trout producer. Models trained above 90% accuracy (95%+ per the 2026-07-26 Launch YC post) and cut inspection time from about 5 minutes to under 30 seconds per fish; two more farm deployments were planned for early 2026 and delta robots for automated sorting were being integrated.

How it ended up

Still running and expanding: two active founders, deploying into more farms and adding robotic sorting as of September 2026.

Background

OctaPulse is a YC W26 startup founded by CMU alumni Paul Grech and Rohan Singh, betting that fish farms will buy computer vision and robotics for the quality checks still done by hand. Most hatcheries net a few dozen fish, anesthetize them and measure them one by one — roughly five minutes per fish — then extrapolate to populations of hundreds of thousands.

The bet is anchored in a paid six-figure pilot with Riverence, North America's largest trout producer: models exceeded 90% accuracy and inspection time dropped from about five minutes to under 30 seconds per fish. The July 2026 Launch YC post puts accuracy at 95%+ and says two more farm deployments were planned for early 2026.

OctaPulse's plan is a ladder: standardize hatchery phenotyping and deformity inspection first, then add delta robots for sorting, then expand to feeding, health monitoring and processing. Each deployment adds labeled images to a proprietary multi-species dataset the founders call the brain for future autonomous aquafarms.

What has to be true

  • Manual inspection is slow (about 5 minutes per fish), technician accuracy can drop below 70% as fatigue sets in, and samples are tiny — so even basic measurements at scale are a step change.
  • Chickens gained decades of selective breeding while most farmed fish remain near-wild genetics; automated measurement is the unlock, and the dataset becomes a moat competitors cannot copy quickly.
  • Starting with drop-in QA sidesteps the trust problem of handling live animals: farms keep their existing workflows while OctaPulse proves accuracy on one large reference customer.
  • The 111-point Launch HN thread and a 6-figure paid contract show demand beyond novelty, which let the company plan two more deployments within its first year.

What can be applied

In an industry that measures almost nothing, the wedge is measurement: sell the drop-in QA step first, let the dataset become the moat, and expand from inspection into sorting and breeding.

Aftermath

As of 2026-09-02, OctaPulse is active with two founders (Grech and Singh) in YC W26. It reported a 6-figure paid contract with Riverence, North America's largest trout producer, model accuracy above 90%, inspection time cut from about 5 minutes to under 30 seconds per fish, two more farm deployments planned for early 2026, and delta-robot sorting integration in progress.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases