NEURAVANT Ibadan, Nigeria

Crop-disease scouting · cassava & maize

The leaf tells you early.
Most farms find out at harvest.

Neuravant is a crop-disease scout for cassava and maize farms in Nigeria: it turns what a grower can see on the leaf into the likely disease, how it travels, and what to do about it this week.

Company
NEURAVANT AI TECHNOLOGIES LIMITED
Ibadan, Oyo State, Nigeria
Team
Oluwafemi Rokeeb Opeyemi, Founder. Sole founder, and the person who writes the code.
Stage
Scout runs in the browser today. The photo model is in training.

The land is not the problem.

Nigeria grows more cassava than anywhere else and still takes under a fifth of what the same hectare could give. What happens between planting and harvest is mostly invisible until it is expensive: a mosaic leaf becomes next season's planting material, a whorl full of frass becomes a thin cob. Someone has to see it in week two.

62m tonnes

of cassava a year — more than any other country on earth, and the crop this starts with. FAOSTAT, 2024

1 : 6,446

farm families to every extension agent, against the FAO-recommended one per thousand. NAERLS, 2024

54,000 ha

lost to pests and disease in the 2024 wet season, with fall armyworm rated moderate to heavy on maize in four states. NAERLS survey

Not here yet

Cassava brown streak can take the whole root crop and has not established in Nigeria. Every scouting record is also a tripwire. Frontiers / IITA

Cassava yield in Nigeria

Modelled across 30 states, the average field is short by 18.2 tonnes a hectare — the same land could give roughly five times what it gives now. Both figures are the paper's modelled dry-matter basis, so they sit below FAOSTAT's fresh-root tonnage. Nutrient and water management carry most of that gap; disease and pest loss is the part a grower can act on in week two. Srivastava et al., Frontiers in Sustainable Food Systems, 2023

~4.1 t/ha observed ~22.3 t/ha modelled potential, same land

Scout. Working, right here.

This is the part of Neuravant that runs today. Pick the crop, tick what you can see, and it ranks the candidates, shows which of your observations drove the ranking, gives you the week's actions and names the one observation that would settle it. It runs in the browser with no signal and sends nothing anywhere.

Tick what you can see in the field. Nothing is uploaded.

What do you see?

Nothing ticked yet. Start with the leaf, then the stem, then cut one root open.

Scout narrows the list and tells you what to check next. It is not a substitute for your extension officer, and it never recommends a chemical or a dose.

Then the phone does the looking.

Today we train the vision models that read a leaf — cassava mosaic, bacterial blight, green-mite damage, maize streak, leaf blight — on cloud NVIDIA L40S GPUs, with CUDA, PyTorch and NVIDIA DALI moving tens of thousands of field photographs through every epoch.

1Public leaf datasetscassava and maize sets releasedfor research, shot in the field2NVIDIA DALIdecode and augmenton the GPU, not the CPU3PyTorch + CUDAtrained on cloudNVIDIA L40S GPUs4Held out, not shuffledscored on photographs thetraining split never saw5Back into Scoutthe photo lands besidethe symptom path, not over it

Today

  • Scout: symptom triage, the week's actions and the next thing to check — offline, in the browser.
  • Vision models in training on cloud NVIDIA L40S GPUs with CUDA and PyTorch.
  • NVIDIA DALI doing decode and augmentation on the GPU, which is what makes a photo-heavy epoch affordable.
  • Scored on photographs the training split never saw, grouped by where they came from — a random shuffle of the same images flatters a model that has simply memorised one place.
  • Training data is public: research datasets of cassava and maize leaves, photographed in real fields.

Next

  • NVIDIA TensorRT for batch checks, so an agribusiness can push a day of photographs through at once.
  • NVIDIA DeepStream so a scout can walk a row with video instead of stopping at every plant.
  • A distilled model small enough to run on the phone itself, for fields with no signal.
  • Spread maps across a catchment, for the extension programmes that need to know what is moving.

The company

Today the scout runs in the browser: symptom triage, what to do this week, and what to check next — offline, on any phone. The photo model is in training. Next: batch checks for agribusinesses and video scouting down a row.

Legal entity

NEURAVANT AI TECHNOLOGIES LIMITED
Ibadan, Oyo State, Nigeria

Who builds it

Founded and built by Oluwafemi Rokeeb Opeyemi, Founder — sole founder, and the person who writes the code.

Who it is for

Cassava and maize growers, and the agribusinesses and public extension programmes that need to know what is spreading in their catchment.

Privacy

Scout keeps everything on the device: no account, no upload, no analytics, no cookies. This page makes no third-party requests at all.

Contact

rokeeb@neuravantlabs.com