Research at TechLab X doesn't sit in papers. It runs inside products that real riders and real farmers rely on every day — and the field is our benchmark.
If a model can't survive real roads and real farms, it isn't done. Every line of research here ends in a shipping product.
Low bandwidth, patchy data, hard infrastructure. We treat those constraints as the problem statement — not an excuse.
Our benchmarks are 200+ farmers and 13 towns of riders — not just held-out test sets.
Convolutional networks that read plant health from a photo — trained in-house on 50,000+ images across 39 crop and disease classes.
Networks of agents that watch farms, reason over live data and act — including triggering real payouts — without a human in the loop.
NDVI and NDWI pipelines over NASA satellite imagery, plus real-time inference on live drone feeds.
The AI-powered engine matching riders to drivers in seconds across 13 towns — balancing wait times, earnings and fairness.
A farmer photographs a leaf; our convolutional network identifies the crop and its disease and returns a treatment plan in seconds. The same model reads live drone feeds frame-by-frame for large farms.
Every six hours, an autonomous agent pulls fresh satellite data for each insured farm. When vegetation health collapses and drought risk turns severe, it executes a smart contract and pays the farmer — no paperwork, no human approval, and every decision recorded on an immutable audit trail.
One oracle cycle
Xryde's AI-powered dispatch engine scores every driver–rider pairing in the city at once, then assigns whole batches together — trading off rider wait time against driver earnings, and protecting new drivers so the marketplace stays fair as it grows.
What the engine weighs on every match
| Model | Approach | Result | Lives in |
|---|---|---|---|
| Crop disease detection | Custom CNN · EfficientNetB3 transfer learning | 98.7% validation accuracy · 39 classes | AgriSense AI |
| Advisory intelligence | LLM advice grounded in NDVI, drought risk & location | Context-aware treatment plans per detection | AgriSense AI |
| Crop yield forecasting | Gradient boosting on climate & input features | Season-level yield estimates per farm | AgriSense AI |
| Smart irrigation | Random-forest classifier on soil & weather sensors | 89.5% decision confidence | AgriSense AI |
A dispatch engine that learns from every completed trip — in development for the next generation of Xryde matching.
Machine perception and control research pointing toward full self-driving on African roads.
Hardware & autonomy →Running our models where the work happens — on vehicles, drones and farm hardware, not just in the cloud.
Hardware & autonomy →
We're a small team shipping research into the real world from Benin City. If that sounds like your kind of lab, come build with us.
