Hatcheries
Sell by count, with proof.
- Count a tray in seconds
- Proof behind every sale
- Report, certificate and invoice from one count
- Batches, quality checks, traceability
For shrimp hatcheries
Every larva.
Counted.
One phone photo of the tray. A count your buyer can trust.
Live 3D demo. Every larva drawn here is counted by the page itself.
Inside one count
From photo to number.
- 01
The photo
One phone photo of the tray. Nothing else.
- 02
Tray found
Walls, rim and glare are left out.
- 03
What the counter sees
Each larva shows up as a dark head spot.
- 04
2,548 head spots
Clumps are split, one spot per larva.
- 05
Every larva marked
Check the marks. Tap to add or remove one.
Counted on this tray
0
larvae. Every one marked on the photo.
Real tray photo from a pilot hatchery, and our counter's real output on it.
15–20 min
per tray by hand, today
Seconds
Counts in seconds, not minutes
0 bars
Works with no signal
0
English, Telugu, Hindi and Tamil
The journey of one count
Hatcheries sit by the sea.
Shed after shed of rearing tanks.
Inside, the water never stops moving.
Every tank is sold by count.
Today that count is done by hand, spoon by spoon.
Thousands in every tank.
Each one counts.
From larva to harvest.
Counted right at the start.
For farmers
Then the seed travels to the pond.
Every pond is stocked by count.
Now the proof arrives with the seed.
Marked photo, count and certificate, on WhatsApp.
A 3D film we built and rendered in Blender, modelled on real hatchery sheds and grow-out ponds.
Who gains
One count. Trusted at both ends.
Farmers
Stock the pond with confidence.
- See the marked photo behind your seed
- Certificate with count, PL stage and date
- No recount at the pond
- Nothing to install
Get the app
A real app on your phone
Install ShrimpCount on Android or iPhone. It opens from your home screen like any app, and keeps working with no signal.
Free. No app store account needed. Try free: 2 counts, no account needed. Works without signal.
On a computer? Scan to open it on your phone.
Guided cameraTap to fix marksReads only the water
Handles crowdingSize by PL stageBatches & averages
Consistency chartCount history + CSVCount report
Delivery certificateInvoicePL orders & payments
Price logPL quality checksDisease tests
TraceabilityProduction logWater quality
InventoryTasksGrowth projection
CalculatorsOffline queueTeam backup
4 languagesBig text
Honest about accuracy
Within 1% of an 1,800 hand count.
We are still testing other PL stages with hatcheries. Until you have compared it with your own hand count on your own trays, treat the number as an estimate.
Pricing
One price. Count as much as you like.
It is free during the pilot.
₹10,000
one-time setup - includes your first month
then ₹5,000 / month after that
- Unlimited counts and photos
- Count report, delivery certificate and invoice
- Backup and sharing across your whole team
- All four languages, on any number of phones
Paid monthly, by UPI or bank transfer - no card on file. Try free first: 2 counts, no account needed.
Questions
Engineers, recruiters and students: the last three are for you.
Does it work without internet?
You can take photos and count by tapping with no signal. The automatic count runs when you are back online.
Who can see my records?
Records stay on your phone. With an account they are backed up to a private space that only your hatchery team can open.
What does it cost?
It is free during the pilot. Then ₹10,000 one-time setup including the first month, and ₹5,000 a month per hatchery.
Do I have to make an account?
No. You can count without one. An account only adds backup and sharing between phones.
How do I install it?
Tap Install on this page. On iPhone, open the app in Safari, tap Share, then Add to Home Screen. No app store needed.
Does it count a whole tank?
No. It counts the sample on a standard tray. A batch of samples gives the tank estimate.
I am a farmer. What do I get?
Your hatchery can send the marked photo, the count report and the delivery certificate with your seed. You install nothing.
How does the counter work?
It finds the tray, removes the background, and looks for the dark head spot of each post-larva, since the body is almost transparent. Noise and faint tails are filtered by size and contrast, clumps are split by expected size for the PL stage, and a person confirms the marks. A classical OpenCV engine runs today; a YOLO11 detector (ONNX) must beat it on the same held-out batches before it ships.
How is accuracy measured?
Data is split by batch, never by random image, so the test set is a day the model has not seen. Only numbers from the evaluation script on held-out batches are quoted. No synthetic images stand in for real hatchery photos.
What is it built with?
An offline-first React PWA, a FastAPI counting service, OpenCV and ONNX Runtime. This page uses three.js and GSAP.