Satellite context
Stored Sentinel-2 observations support vegetation, moisture-context and crop-canopy trends. Cloud and valid-pixel quality checks come before an inspection prompt. A changed boundary makes earlier evidence historical.
THE EVIDENCE BEHIND THE EXPERIENCE
More context. Clearer limits.
Different sources answer different questions. Krishyak keeps their dates, methods and limitations visible, so a useful clue is never presented as a certainty.
Explore the demo
Every source has a job.
Weather and soil
Weather is modelled context for an approximate field location. Soil evidence keeps its reported method, date, depth and units. Krishyak does not infer soil nutrients from satellite colours.
Farmer observations
Your field notes and recorded actions stay attached to the field and crop season. They are farmer-entered evidence, not automatically verified measurements.
A crop photo is a starting point.
For a supported crop, the classifier suggests an image label and records a raw model score. It does not measure severity, prove a field diagnosis or choose a chemical dose. Farmers can mark a result useful, incorrect or uncertain, and research review follows separate consent and role checks.
About the current model and its validation
38 labels across 14 crop categories
The active release is publisher-efficientnet-colab-v1, using an EfficientNet classifier and a verified compact inference artifact.
Recorded internal test accuracy is 93.71% on 8,566 examples. External PlantDoc accuracy is 57.21% on 229 examples. That external result is a material limitation for field photographs; neither number establishes local field performance.
The 95% Wilson interval for the external result is approximately 50.73%–63.44%. Real field validation and label-by-label agronomic review remain required.
Read model governancePrivate records, by design.
Account ownership
Farm and plot records are checked against the signed-in account. Opaque sessions, origin checks and request quotas remain in the existing FastAPI backend.
Private photographs
Uploaded images are validated, normalized and stored behind authenticated access. Metadata export does not expose private object keys.
Purpose-based permissions
Location, satellite processing, agronomic analysis and optional research purposes are separate controls. Withdrawing permission affects future processing.
Your next step starts with a clearer picture.
Take a look around a demonstration farm, or start keeping your own field records when farm accounts are available.