How Abelio Reduced Time-to-Model for Precision Agriculture

From fragmented AWS infrastructure to unified MLOps: Abelio now delivers farmer insights within 48 hours of image acquisition.

48h
Retraining Cycle
Image to insights
4x
Seasonal Scaling
Peak processing increase
TBs
Data Managed
Dozens of terabytes
Company
Abelio
Agriculture

Overview

Abelio is a digital farming solutions provider that uses computer vision to process aerial imagery from drones and satellites. They deliver insights that help farmers optimize yields, reduce costs, and minimize environmental impact across large-scale agricultural operations.

01 — The Challenge

Abelio needed to process massive volumes of aerial imagery during peak farming seasons while meeting strict 48-hour delivery timelines.

  • Massive Data Volumes: During peak farming seasons, image processing increased fourfold. Managing terabytes of drone and satellite imagery required robust infrastructure.
  • Tight Delivery Timelines: Farmers needed insights within 48 hours of image acquisition. Time-consuming model retraining threatened these critical deadlines.
  • Fragmented Infrastructure: AWS services (S3, EC2, SageMaker) lacked integrated tools for image visualization and dataset management, creating workflow gaps.
  • No Reproducibility: Lack of traceability across diverse environments made it impossible to reproduce experiments or track what worked.

02 — The Transformation

From challenges to solutions

Before Picsellia

  • Fragmented AWS infrastructure (S3, EC2, SageMaker)
  • No integrated tools for image visualization
  • Time-consuming model retraining processes
  • Lack of traceability across environments
  • Difficult to reproduce experiments

After Picsellia

  • Centralized data management platform
  • Efficient retraining workflows
  • Full experiment reproducibility
  • Streamlined annotation campaigns
  • 48-hour delivery to farmers

“With Picsellia, we can now deliver insights to farmers within 48 hours of image acquisition. The platform handles our seasonal data spikes without issues while maintaining full traceability.”
A
Abelio Team
Data Science, Abelio

03 — The Workflow

How Abelio uses Picsellia

01
Ingest Aerial Imagery
Centralize drone and satellite images with metadata into searchable datalake

02
Run Annotation Campaigns
Structured campaigns with progress tracking and quality control for crop analysis

03
Rapid Model Retraining
Fast iteration cycles with versioned datasets and tracked parameters

04
Monitor Model Performance
Track accuracy across different crop types and seasonal conditions

04 — The Solution

How Picsellia delivered

Picsellia provided Abelio with a unified platform to manage their entire agricultural imaging pipeline at scale.

Centralized Data Management

Simplified image storage and organization for terabytes of aerial imagery. Powerful querying and visualization capabilities.

TBs
Images managed

Efficient Retraining Workflows

Streamlined processes to meet 48-hour delivery timelines. From image acquisition to farmer insights in record time.

48h
Delivery time

Annotation Quality Control

Campaign tools enabling progress tracking and quality control. Improved annotation efficiency and consistency.

Full Reproducibility

Dataset versioning and parameter recording ensures every experiment can be reproduced and compared.

100%
Reproducibility

05 — The Results

Business impact

Abelio transformed their agricultural imaging pipeline to deliver faster, more reliable insights to farmers.

  • 48-Hour Delivery: Reduced retraining cycles to meet strict farmer delivery timelines, from image acquisition to actionable insights.
  • Improved Model Accuracy: Better dataset reliability stabilized models, increasing precision and recall across crop analysis tasks.
  • Seasonal Scalability: Effectively managed fourfold increases in data volume during peak farming seasons.
  • Enhanced Reproducibility: Full experiment tracking through dataset versioning and parameter recording.

Picsellia Features Used

Datalake
Large-scale image storage

Annotation Campaigns
Quality-controlled labeling

Experiment Tracking
Reproducible training

Model Monitoring
Performance tracking