AI for aquaculture operations

Count the seed.
In murky water.

I point a camera at the tank and count 10mm seed through turbid water so the biomass estimate stops depending on hand samples.

No login
No IT
No pilot
Tank cameraseed count
COUNTED
live count HUD
flagged: 10mm seed counted through murky water
pain

Hand counting

Manual counting gets accepted as just how it is done.

pain

Sample estimates

Biomass gets extrapolated from a small oyster sample everyone knows is off.

pain

Murky water

Normal cameras fail when the work happens in turbid water.

pain

Traceability pressure

Buyer traceability demands are getting stricter, and the count still starts by hand.

What I build

A camera counter for the tank.

Your hatchery team still owns husbandry decisions. My system does the watching: seed count, size estimates, object separation, frame review, and traceability events that need clean evidence.

01

Point at the tank

I start from the camera angle you can actually use around water, glare, and moving seed.

02

Find each seed

The system separates tiny oysters from murk, motion, shell, and background noise.

03

Count and size

Each detected seed gets counted and measured so the estimate is visible, not guessed from a sample.

04

Review the frames

Operators can inspect the flagged frames before any count changes a production decision.

Seed-count sharp tip

The hand sample gets replaced by visible detections.

Vision path
Capture the tank or tray.
Detect and size individual seed.
Produce a reviewable count with flagged frames.

Secondary pass: traceability mapping for critical tracking events where buyers need clean records.

Receipts

The CV proof is literal.

aquaculture vision
10mm

Computer vision that counts 10mm oyster seed in murky water.

validated benchmark
100%

100% statute recall on a validated benchmark.

The same operator who builds deterministic legal systems built computer vision that counts seed in the water where the work actually happens.

The Diagnostic

How far off is your biomass estimate when it's extrapolated from a 30-oyster sample?

Fixed-scope Diagnostic. In 14 days I map the counting workflow, camera constraints, sample logic, and the narrow place where vision can remove the guess.

I do not promise yield, mortality, or food-safety certification outcomes. A human still reviews the count and the operational decision. My system flags and extracts so the human is faster and misses less.

Start with the Diagnostic.

$950, fixed. Send what you already have — the messy spreadsheets are the data — and get back the one gap producing your thirty symptoms. 2–4 business days. Credited toward the first build if we continue. Yours either way.