The bribery loophole
Local inspectors face pressure from Kolar Arhatiyas to sample only the good crates, or to tilt the phone and distort the size calibration.
Kolar Mandi · Tomato Grading
Grade the whole crate, not just the top layer.
We put fair, anti-bias grading into the hands of mobile inspectors on the mandi floor. A statistical core-sample and on-device AI deliver a trusted grade in under 45 seconds — no conveyor, no internet, no million-rupee dumper.
- Per-lot grade
- 0
- Per inspector kit
- ₹0
- Handling damage
- 0bruise
01 The Problem
The best tomatoes always sit on top.
A 26 kg crate holds roughly 1,500 tomatoes. Passive, top-down scanning — and the human eye — only ever read the surface. Sellers know it, and they pack the surface accordingly. The grade looks honest; the crate is not.
Fixing this with industrial dumpers and fixed conveyor lines costs millions of rupees per mandi — capital a startup can’t justify. So we changed where and how the grading happens, not how much steel it takes.
You don’t need to scan all 1,500 tomatoes to know the truth — only a fair, random sample of them.
02 The Method
The Core Sample inspection
An active-sampling, mobile-first workflow that mixes top, middle and bottom into a single honest reading. Start to finish in under 45 seconds.
-
01
Random selection
For a lot of 100 crates, the app flags 3 crates at random — and only after the inspector has logged into that specific lot.
-
02
The funnel dump
The crate is tipped into a matte-black funnel that gently rolls every tomato into a single-layer tray — bottom, middle and top, fully mixed.
-
03
The single shot
One photo, held parallel over the tray. The model isolates every tomato and reads colour, size and blemish in a single frame.
-
04
Seamless repack
The tray tilts and the tomatoes slide back into the crate. Zero manual sorting, zero handling damage.
03 Technology
A lab inside a ₹15,000 phone
No mini-PCs, no industrial cameras, no variable drives. The whole grading engine runs YOLOv8-nano / MobileNetV4 through TensorFlow Lite, locally on the handset — fully offline on the dusty mandi floor.
Runs on-device · no signal required-
A
Colour matrix
Every tomato in the frame is isolated and read for its average HSV profile, returning a live ripeness breakdown.
-
B
Size calibration
A 30 mm blue disc is fixed to the tray as a geometric anchor. No fixed camera height needed — the app scales from the disc.
-
C
Blemish & puncture
Trained to flag localised anomalies on the tumbled fruit — black rot, white fungal patches and open skin tears.
04 The Kit
Ten inspectors for the price of one conveyor line
The entire anti-bias grading rig is cheap to replicate — so coverage scales with people, not capital.
ComponentWhat it isCost (INR)
Mobile app license
In-house white-label CV app running local TFLite models.
₹0scale cost
Smartphone
Android handset, 48 MP camera, mid-range processor.
₹15,000
Sampling rig
Folding aluminium frame, matte funnel, flat tray, phone mount.
₹8,000
Calibration target
High-durability fixed-dimension reference token.
₹200
Total per inspector
A fully mobile, anti-bias grading kit.
~₹23,200 ≈ $280
05 Integrity
Engineered against its own weak points
The human element is where systems break. We stress-tested ours and built the defences in.
Perishable damage
Tomatoes bruise easily. Aggressive dumping causes micro-fractures that accelerate rot across the 36-hour journey to Delhi or Kolkata.
06 Get started
Bring fair grading to your mandi.
Run a pilot with Honour Agriculture — deploy a mobile grading kit on your floor and see the difference between the top layer and the truth.