Speed is the most critical variable in crop disease management. A disease identified in its first week can almost always be controlled. A disease identified in its third week has frequently spread beyond effective intervention. The gap between those two outcomes is not knowledge — most diseases have known treatments — it is access to accurate diagnosis at the moment it matters.
ARCORA is built around that constraint. The design question was never "how can we diagnose crop diseases" — the science for that has existed for decades. The question was "how can we put accurate diagnosis in the hands of a farmer in a field, in under two minutes, using only a photograph."
Here is what the answer looks like.
The Input: A Photograph of Any Affected Part
The farmer opens ARCORA on their phone and photographs the affected plant. The system accepts images of any affected part: leaves, stems, fruits, flowers, or roots. This matters more than it might seem.
Most crop disease diagnostic tools are trained primarily on leaf images because leaves are the most visually distinctive and the most extensively photographed in research datasets. But diseases frequently present earlier, and more diagnostically clearly, in other parts of the plant. A root rot that has not yet reached the leaves. A stem borer that is visible in the stem before the canopy shows distress. A fruit disease that a farmer notices while harvesting before the leaves have shown any symptoms.
Restricting analysis to leaves would make the tool simpler. It would also make it systematically less useful for the kind of early diagnosis that actually changes outcomes. ARCORA accepts any part because farmers encounter disease in any part.
The Analysis: Multimodal Visual Intelligence
The image is processed by a diagnostic model trained on hundreds of thousands of annotated crop disease images across the major crops grown in Karnataka and the broader Indian agricultural region. The model identifies the visual signatures associated with specific diseases — the pattern, colour, morphology, and distribution of symptoms that distinguish one condition from another.
This is more difficult than it sounds. Many crop diseases produce similar visual presentations in their early stages. Early blight and late blight in tomatoes look nearly identical to an untrained observer and respond to entirely different treatments. Nutrient deficiency and fungal infection produce similar leaf discolouration patterns. Bacterial and viral infections may present with overlapping symptoms before their distinctive features develop.
The diagnostic model is built to resolve these ambiguities — to identify the specific combination of visual features that distinguishes between conditions that look similar. Where the confidence threshold for a primary diagnosis is not met, the system produces a differential and identifies which additional information would resolve it.
The Output: A Complete Treatment Protocol
Within two minutes of the photograph being submitted, the farmer receives not just a diagnosis but a full treatment protocol.
The protocol includes: the name of the disease, both common and scientific; the causal agent, whether fungal, bacterial, viral, or nutritional; the recommended treatment with specific product names, active ingredients, dosage, and application method; the timing and frequency of treatment; and any additional agronomic interventions that improve outcomes.
This completeness is deliberate. A diagnosis without a treatment protocol places the burden back on the farmer to figure out what to do with the information — to find a dealer, to ask about products, to hope the advice they receive is correct. The protocol closes that loop. The farmer knows what to buy and what to do before they leave the field.
The output is provided in Kannada as well as English, because a diagnosis in a language the farmer cannot comfortably read is not a diagnosis — it is an obstacle.
The Network Effect: How Every Diagnosis Makes the System Smarter
Each diagnosis submitted through ARCORA — the image, the confirmed identification, the treatment outcome where it is reported — feeds back into the system's knowledge base. This creates a compounding effect that standard diagnostic tools do not have.
After five hundred tomato disease diagnoses in Kolar, the system has a richer understanding of the disease patterns, timing, and progression specific to that region's conditions than any general agricultural research database. It knows which diseases peak in which seasons in that microclimate. It knows which varieties are more susceptible to which conditions in that soil type.
This locality-specific intelligence is valuable in a way that general agricultural knowledge is not. A farmer in Kolar is not growing tomatoes in average Indian conditions. They are growing them in specific soil, in a specific climate, at a specific elevation, with specific inputs. A diagnosis system that has learned from five hundred previous cases in that specific context is materially better than one that hasn't.
What ARCORA Does Not Do
ARCORA is a diagnostic and advisory system. It is not a prescription system. It provides treatment recommendations that the farmer can discuss with an agronomist or act on with appropriate guidance. It does not replace expert agricultural advice for complex cases. For disease presentations that are unusual, severe, or do not match the training data's patterns clearly, the system flags this and recommends consultation with an agricultural specialist.
The honesty of this limitation is important. A tool that claims to solve every agricultural problem creates the kind of overconfidence that leads to bad decisions. ARCORA is built to be maximally useful in the large majority of cases where fast, accurate identification changes outcomes — and appropriately humble in the minority where human expertise is still the best resource.
That balance — confident where confidence is warranted, cautious where it is not — is what makes a diagnostic system trustworthy over time.