Skip to content
AtheronLABS

You're visiting from the United States. Prices are shown in US dollars. Not right?

labs@atheron:~/industries/agriculture/ai$ train agent --grounded

AI crop scouting

Custom vision models trained on your own fields.

A general image model knows what a leaf looks like; it does not know your crops, your diseases or your region. A custom model fine-tuned on images your agronomists label does, and it improves each season as their corrections are fed back into training.

// what matters here

What is different in this industry.

01

Trained on your fields

The model learns from photos taken on your farms and labelled by your agronomists, not from a stock set gathered somewhere else.

02

A suggestion, not a diagnosis

The model says what it sees and how sure it is; the agronomist confirms or corrects, and the decision to spray stays with them.

03

Measured before each season

Each new version is scored against a held-out set of your images, and it is used only if it does better than the last.

04

Your photos stay yours

Images and the trained model are kept in your own cloud account, and you own both once paid for.

// example projects

Priced examples for this service.

Each one is hypothetical, labelled as such, and priced live from our rate card. Open any of them in the estimator and make it yours.

Example projectWeb appMobile appAICustom LLM

Crop scouting with custom vision

Hypothetical. Not a client, not a result.

// The problem

An agronomy service scouts many fields every week, and catching disease and weeds early depends on which agronomist walked the field that day.

// What we would build

A scouting app where agronomists and growers photograph leaves and plants, with a custom vision model fine-tuned on your own labelled images of local crops, diseases and weeds suggesting what it sees and how sure it is. A custom model fits because general models do not know your crops or your region; it runs on GPUs you rent, and the photos stay yours.

// What is in it

  • Phone photos tagged to the field and block they were taken in
  • A model fine-tuned on your labelled images, measured on a held-out set before use
  • Suggestions with a confidence level, confirmed or corrected by the agronomist
  • Corrections fed into the next round of training
  • Maps of findings by field across the season
  • Photos queued on the phone until it has a signal

// Stack

  • PyTorch
  • React Native
  • Python
  • Next.js
  • PostgreSQL with PostGIS
  • AWS GPU

// estimate

Build
≈ US$171,000 to US$261,000, delivered within 33 weeksCAD 243,600 to 372,000
Hosting
≈ US$6,710 a monthCAD 9,550 a month
Support
≈ US$2,500 a monthCAD 3,565 a month

Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.

AI route
Custom LLM
Model running cost
≈ US$6,530 a monthCAD 9,300 a month

Timeline by milestone

Discovery
2.9 to 3.4 weeks
Data and evaluation plan
0.4 weeks
Design approved
3.4 to 5.9 weeks
Core features
1.4 to 2.4 weeks
Full build
0.9 to 1.4 weeks
First build on devices
2.4 to 4.4 weeks
Training
4.9 to 8.4 weeks
Testing and fixes
0.4 weeks
Store submission
0.4 weeks
Evaluation and red-teaming
3.9 to 4.4 weeks
Launch
0.9 to 1.4 weeks
Production
0.9 to 1.4 weeks

Outside our hands, and added to the calendar

App store review
1 week

How it is paid

Deposit 20%
CAD 48,480 to 74,160
Discovery 4%
CAD 9,696 to 14,832
Data and evaluation plan 5.6%
CAD 13,574.40 to 20,764.80
Design approved 4%
CAD 9,696 to 14,832
Core features 3.3%
CAD 7,999.20 to 12,236.40
Full build 2.2%
CAD 5,332.80 to 8,157.60
First build on devices 8.7%
CAD 21,088.80 to 32,259.60
Training 16.9%
CAD 40,965.60 to 62,665.20
Testing and fixes 1.1%
CAD 2,666.40 to 4,078.80
Store submission 2.9%
CAD 7,029.60 to 10,753.20
Evaluation and red-teaming 11.3%
CAD 27,391.20 to 41,900.40
Launch 4%
CAD 9,696 to 14,832
Production 6%
CAD 14,544 to 22,248
Holdback, 30 days after launch (10%)
CAD 24,240 to 37,080
GPU time for training and testing, at cost
CAD 1,200. Billed up front, at cost, outside the milestones

// questions

Questions we get about this.

How many labelled photos does a crop model need?

It depends on how many crops and problems you want covered. Fine-tuning often starts from a few hundred labelled photos per problem; we measure as we go and tell you when more would help.

Can the model run on the phone without a signal?

A smaller version can, for the most common problems, with the full model checking photos when the phone reconnects. The estimate shows both options.

Why custom rather than a frontier model for scouting?

Frontier models describe photos well in general but are not reliable on local diseases and weed species. A fine-tuned model is more dependable on the problems it was trained on, and it costs less to run at volume.

// next

Not quite your project?

Tell us what you have in mind. We will come back to you with a range and the questions that would narrow it. Or book a call and talk it through.

Custom AI vision models for crop scouting | Atheron Network Labs