Neuralis

A reform announcement can matter to cocoa farmers, but it does not tell a farmer standing beside a damaged tree what to do next. Before action, that farmer still needs a safe answer in the language they use at work, with a clear route to a person when the answer is uncertain.

At 6:40 in the morning, Kojo stands at the edge of his cocoa plot outside Kumasi with a pod in one hand and his phone in the other. The pod has a dark patch near its stem. Rain sat on the leaves overnight, and he has already heard a radio bulletin about cocoa reform while making tea.

The bulletin gave him the bigger picture. It did not tell him whether this pod points to a disease, whether he should remove it, or whether a chemical product in his shed is safe to use. His neighbour is waiting for an answer before they start work. If Kojo guesses wrong, the problem could spread through more of the plot, or workers could be exposed to an unsafe spray.

That is the gap between an announcement and a decision.

A farmer acts at the tree, not at the radio

News about cocoa policy reaches people through radio, conversations, cooperatives and local meetings. Farmers in parts of the Ashanti Region have welcomed the passage of the Ghana Cocoa Board Bill, 2026. That response makes sense. Changes to the institutions around cocoa affect work that supports homes, school costs and the next farming season.

Yet broad information has a different job from immediate guidance.

A farmer may hear that reform is coming and still face a pod with an unfamiliar mark that afternoon. They may need to ask in Asante Twi, using the words that come naturally in the field. They may mix Twi with English product names. They may have a noisy voice note, a torn label, or a question with missing details.

The useful response has to meet that moment. It must speak plainly enough to understand, stay within advice that a reviewer has approved, and avoid filling gaps with confident language.

The safest answer can be “speak to the extension officer”

Kojo records a short voice note. He describes the dark patch and asks whether he should spray. A voice system can help only if it first recognizes what it knows and what it does not.

For a familiar, reviewed question, it can select a pre-approved guidance block and return it in spoken Twi. For an unclear report, it should ask for the detail needed to narrow the problem. For a pesticide question without verified dosage, re-entry or pre-harvest guidance, it should stop and escalate.

That pause can feel unsatisfying when rain is close and the workday is moving. It is still better than a polished answer built on a missing safety fact.

Neuralis is preparing AgriVoice around that boundary. The system is designed to select reviewed content rather than invent agronomy advice. Chemical responses retain mandatory safety content, and uncertain cases must reach a named human owner. The point is practical: farmers should never receive a fluent instruction that nobody has checked.

The risk is real even before chemicals enter the conversation. Translation can silently distort the noun that carries the whole question. In one test, the Twi word “kokoo,” meaning cocoa, came back as “chicken.” A response built on that switch may sound coherent while addressing the wrong crop entirely. Agricultural Voice AI: Why Kokoo Becoming Chicken Must Trigger Human Review explores why that kind of error needs a human review path.

A voice workflow has to earn trust in the field

Back at the tree, Kojo does not need a long explanation of artificial intelligence. He needs to know what the next safe move is.

If the system recognizes the question and has reviewed guidance, it should answer quickly in language he can follow. If the question is outside its reviewed material, it should say so clearly and pass the case to the extension officer responsible for escalations. A vague “please try again” leaves the burden where it started. A specific handoff gives the farmer a path.

That is why the AgriVoice pilot is built as a learning gate, not a public scale launch. A cocoa-sector partner must own recruitment and name the extension officer receiving escalations. The spoken Asante Twi content must be reviewed by someone who understands farming vocabulary. A Ghanaian agronomist must clear the held chemical guidance. Staff must rehearse the full path before farmer exposure.

The pilot also has to measure whether people understand the voice, ask a second question, receive answers or correct escalations, and get help fast enough when the system cannot answer safely. A response that arrives but is misunderstood has not done its job.

The decision begins with a small, accountable next step

Kojo’s question does not need to become a test of whether a machine can sound certain. It needs an answer that helps him protect the tree in front of him.

If the details match a reviewed disease response, he hears the next step and checks the affected pod again. If the question turns on an unverified chemical instruction, the system does not tell him to improvise. It sends the case onward while the sprayer stays closed.

By mid-morning, the radio bulletin is still part of the day. It gives Kojo a reason to pay attention to what may change around cocoa. The safer voice response gives him something more immediate: a decision he can act on, or a clear reason to wait for the person who can verify it.

Neuralis

AgriVoice helps Asante-Twi-speaking cocoa farmers ask farming questions by voice and receive answers assembled only from agronomist-reviewed content, with human escalation when the system is unsure.

Try Neuralis

Comments

No comments yet.