Mavuno Intelligence · Agricultural AI Platform

AI that reads the field
before you lose the harvest.

54,000+ labeled crop images. Sentinel-2 satellite coverage. Three precision modules — trained on validated agronomic datasets and deployable across every layer of African agricultural value chains.

View Plans Explore Modules

40%

Of harvests lost
to disease yearly

AI modules,
one platform

93%

Mean AUC-ROC
crop disease

10m

Satellite spatial
resolution

What it is

Precision intelligence
for the African field.

Most crop infections are invisible until 30–60% of the harvest is already compromised. At that point, treatment limits damage — but the loss is locked in. BAKAP DataLabs has built a three-module AI platform that detects disease from a smartphone photo in under 500ms, monitors field basins from orbit weekly, and scores agricultural loan risk with agronomist precision.

Trained on PlantVillage, the CGIAR-IITA African crop disease database, and ESA Sentinel-2 imagery — Mavuno Intelligence is the first AI platform built for the scale, the crops, and the economic stakes of sub-Saharan African agriculture.

Get access → See the 3 modules

ANTHRACNOSE 0.91

RUST 0.74

HEALTHY 0.97

Anthracnose · 0.91

Leaf Rust · 0.74

487ms inference

Disease zone flagged

Rust — secondary

Healthy tissue

AI Modules

Three specialized engines,
one unified platform.

From the individual plant leaf to the satellite basin to the financing desk — each module addresses a distinct layer of agricultural intelligence, deployable independently or combined.

Module 01 · Disease Detection

CropScan

Multi-label classification of crop leaves and fruits from smartphone photos. Detects the five highest-prevalence plant diseases with Grad-CAM heatmap localization. No lab. CPU-only. Offline mode available.

AnthracnoseLeaf BlightMosaic VirusLeaf RustRoot Rot

PlantVillage + CGIAR-IITA · AUC ≥ 0.93

Module 02 · Satellite Monitoring

FieldWatch

Automated Sentinel-2 multispectral processing at 10m resolution. NDVI, NDWI, and EVI per parcel — updated weekly. Drought stress, flood detection, and yield trend alerts for basins up to 500,000 ha.

NDVI MappingDrought StressFlood DetectionYield Forecast

ESA Copernicus Sentinel-2 · 10m · 5-day revisit

Module 03 · Risk Scoring

YieldScore

An automated agricultural risk score (0–100) per parcel, fusing CropScan alerts, FieldWatch satellite indices, and CHIRPS rainfall anomalies. Built for loan officers, insurance actuaries, and compliance teams.

Loan Risk ScoreInsurance TriggerClimate RiskHarvest Forecast

CropScan + FieldWatch + CHIRPS · R² 0.87

Who it's for

Six sectors.
One irreplaceable asset.

Agribusiness / Cooperative

"What is the real health of my cocoa basin right now?"

FieldWatch monitors your supply basin weekly. Any parcel with NDVI stress drop above 15% triggers an automatic geo-alert — weeks before harvest loss becomes irreversible. No field visit required for the initial triage.

Value: Early warning saves 20–35% of at-risk harvest

Development Bank / MFI

"Does this agricultural loan carry real harvest default risk?"

YieldScore generates a pre-disbursement parcel risk report: disease exposure, rainfall forecast delta, NDVI trend, and historical yield — compiled into a single 0–100 index for every loan in the portfolio.

Value: Agricultural NPL reduced by up to 40%

Agricultural Insurer

"Did this farmer actually suffer a verifiable climate loss?"

Index insurance claims validated automatically. FieldWatch compares NDVI trajectory before and after the declared loss against historical baseline — producing an audit-ready satellite evidence report. No field adjuster needed.

Value: Claims assessment cost cut by 65%

Export Company (Cocoa / Coffee)

"Will my sourcing zones deliver contracted quality this season?"

Cross your origin network against CropScan disease prevalence and FieldWatch seasonal forecasts. Identify high-risk origin zones 6–10 weeks before the harvest window and reallocate procurement before the quality penalty is written.

Value: Export quality penalties eliminated

NGO / Food Security Program

"Which zones are heading toward a food crisis in the next 90 days?"

Aggregate FieldWatch NDVI with rainfall anomaly data across entire provinces. Weekly predictive food insecurity maps — enabling pre-positioning of aid before a crisis becomes visible on the ground.

Value: 90-day early warning for food emergencies

Input Company (Seeds / Fertilizers)

"Where do my products generate measurable yield improvements?"

Cross your sales territory against FieldWatch yield performance maps to identify zones where your inputs demonstrably outperform competitors. Generate evidence-based agronomic performance reports for every distributor meeting.

Value: Territory performance evidence at field scale

The Invisible Threat

Every undetected infection
is a preventable loss.

The agronomist shortage means one specialist often covers 5,000+ farmers. Most crop disease goes undiagnosed until 30–50% leaf coverage — at which point the yield loss is already irreversible. Mavuno Intelligence gives every cooperative, loan officer, and field agent a diagnostic system that detects infection before the eye does.

Critical loss

Disease at late stage. 40–60% yield loss locked in. Treatment limits spread only.

Moderate loss

Visible symptom stage. 15–35% yield loss. Treatment recovers most of the harvest.

Early detection

CropScan flags infection pre-symptom. Targeted treatment. Yield loss under 5%.

Yield retained per crop — with vs. without early detection

Cocoa

38% — undetected

Coffee

44% — late detect

Maize

52% — moderate

Cassava

41% — undetected

+ Mavuno

94% — early detect

AI Architecture

State-of-the-art models,
field-validated.

CropScan runs EfficientNet-B4 fine-tuned on PlantVillage and CGIAR-IITA. FieldWatch processes Sentinel-2 Level-2A surface reflectance through a custom spectral pipeline. YieldScore fuses both with CHIRPS rainfall anomaly data in a gradient-boosted risk model.

01Image IngestionResize · CLAHE normalize · Multi-scale augment

02Feature ExtractionEfficientNet-B4 backbone · ImageNet pretrained

03Multi-label HeadBCEWithLogitsLoss · Sigmoid · 5-class output

04Spectral FusionB4/B8/B11 Sentinel-2 · NDVI · NDWI · EVI

05Risk Scoring APIJSON YieldScore · GeoJSON parcels · PDF export

# CropScan · EfficientNet-B4
Model(
  backbone="efficientnet_b4",
  pretrained=True,
  num_classes=5, # Anth·Blght·Msc·Rust·Rot
  dropout=0.4
)

# FieldWatch · Sentinel-2 Pipeline
ndvi = (B8 - B4) / (B8 + B4)
ndwi = (B3 - B8) / (B3 + B8)
evi = 2.5 * (B8 - B4) / (B8 + 6*B4 - 7.5*B2 + 1)

# YieldScore · Gradient Boosting
score = GradientBoostingRegressor(
  features=[ndvi, ndwi, cropscan_risk,
           chirps_anomaly, hist_yield_3yr]
)

Training Data

Four authoritative datasets,
rigorously validated.

Every model traces its training data to peer-reviewed, publicly auditable sources. No synthetic data. Full scientific reproducibility.

01 / DISEASE DETECTION

PlantVillage

Penn State University · USA

54,306 images across 38 plant-disease categories from 14 crop species. Expert-annotated under controlled and field conditions. The global benchmark for agricultural AI.

54,306 images38 categories14 crop speciesExpert-labeled

02 / AFRICAN CROPS

CGIAR-IITA

Int'l Institute of Tropical Agriculture · Nigeria

African-context crop disease imagery from Nigeria, Tanzania, Uganda, and DR Congo. Covers cassava mosaic, maize streak, and groundnut rosette — the three highest-impact diseases in sub-Saharan Africa.

Africa-specificCassava · MaizeField-collected4 countries

03 · 04 / SATELLITE + RAINFALL

Sentinel-2 + CHIRPS

ESA Copernicus · UCSB Climate Hazards Group

Sentinel-2 provides 10m multispectral imagery every 5 days. CHIRPS v2.0 supplies 35+ years of daily rainfall records at 0.05° resolution across Africa — feeding YieldScore's climate risk component directly.

10m resolution5-day revisit35yr rainfallOpen access

Validated Performance

Metrics that meet
agronomic standards.

All metrics on held-out test sets with stratified splits. Benchmarked against PlantVillage published baselines and FAOSTAT ground-truth yield survey data.

0.93

AUC-ROC · Disease Detection

CropScan on PlantVillage + CGIAR test set

0.91

AUC-ROC · Crop Stress

FieldWatch NDVI anomaly detection

0.87

R² · Yield Forecast

YieldScore vs. ground-truth harvest records

487ms

Avg. Inference Time

Per photo on CPU · Smartphone-compatible

Pricing

Pricing on request.
Tailored to your scale.

Every plan includes full dashboard access. Pricing is tailored to your basin size, number of users, and deployment model. We respond within one business day.

Starter

Cooperative

From — contact us

up to 3 users

CropScan disease detection module
Up to 1,000 photo inferences / month
Grad-CAM heatmap output per image
PDF diagnostic report per analysis
Web dashboard + offline mobile mode
Email support

Get Started

AI that reads the fieldbefore you lose the harvest.

Precision intelligencefor the African field.

Three specialized engines,one unified platform.

Six sectors.One irreplaceable asset.

Every undetected infectionis a preventable loss.

State-of-the-art models,field-validated.

Four authoritative datasets,rigorously validated.

Metrics that meetagronomic standards.

Pricing on request.Tailored to your scale.

Ready to deployagricultural intelligence?