AgriLumens — Precision Agriculture Platform

Stack

Impact

• Estimates inorganic nitrogen and organic matter from aerial imagery, cutting the dependency on slow, per-sample laboratory analysis.
• Turns a manual research workflow into a repeatable pipeline: the same flight can be reprocessed with a different model without re-flying the field.
• Supported the comparative study of spatial resolutions (GSD) by processing flights at different altitudes through an identical pipeline.
• Designed for reproducibility so later academic work can extend or reuse individual services without redesigning the system.

What I did

• Split the system into independent domain services — gateway, users, processing, models, reports, status — each with its own responsibility and lifecycle.
• Integrated OpenDroneMap (NodeODM) as an external photogrammetry engine to generate orthomosaics and digital elevation models from raw flights.
• Built the ML module as a comparison harness: MLP, SVR and ensemble regressors evaluated with cross-validation on R², RMSE and MAE.
• Modeled the storage layer to keep raw data, processed products and analytical results separate, so experiments can be repeated without losing the originals.
• Made every run traceable — processing date, flight configuration, GSD, model used and code version are recorded alongside the result.
• Containerized the whole platform with Docker Compose and multi-architecture builds so it runs on a laptop or on a Linux server unchanged.
• Wrote the technical documentation set (architecture, data flow, ML, storage, deployment, limitations, reproducibility) that accompanies the research.

Tags