SGS

How SGS Slashed Model Development Time by 66%

From fragmented tools to unified MLOps: SGS transformed their computer vision workflow and went from 3 days to 1 day per model iteration.

Get Similar Results
66%
Faster Development
Per model iteration
3→1
Days per Model
Development cycle
10+
Active Projects
Running simultaneously
Company
SGS
Testing, Inspection & Certification
www.sgs.com
Overview

SGS is the world's leading testing, inspection, and certification company. With over 99,000 employees in 2,600 offices and laboratories, they provide critical services across agriculture, mining, oil, gas, chemicals, and consumer goods. Their computer vision team develops AI solutions to automate visual inspection processes at scale.

01 — The Challenge

SGS's computer vision team was building inspection models across multiple industries, but their workflow was fragmented across disconnected tools. This created significant bottlenecks in their development process.

Fragmented Toolchain

The team was juggling multiple disconnected tools for data management, annotation, training, and deployment. Context switching between tools wasted valuable engineering time.

No Version Control for Data

Without proper dataset versioning, it was impossible to track which data was used for which model. Reproducing experiments or rolling back was a manual nightmare.

Collaboration Bottlenecks

With 4 engineers managing 10+ concurrent projects, there was no centralized way to share datasets, models, or experiment results. Knowledge was siloed.

Slow Iteration Cycles

Each model iteration took approximately 3 days from data preparation to evaluation. This slow feedback loop significantly delayed project delivery.

02 — The Transformation

From challenges to solutions

Before Picsellia
  • 3 days per model iteration
  • Disconnected tools for each step
  • No dataset version control
  • Manual experiment tracking
  • Knowledge siloed between team members
  • Reproducibility was a nightmare
After Picsellia
  • 1 day per model iteration
  • Single platform for entire workflow
  • Git-like versioning for all data
  • Automatic experiment logging
  • Shared workspaces and collaboration
  • 100% reproducible experiments
Picsellia unified our entire workflow. What used to take us 3 days now takes 1. We can finally focus on solving computer vision problems instead of wrestling with our toolchain.
T
Tristan Cotte
Computer Vision Engineer, SGS
03 — The Workflow

How SGS uses Picsellia

01
Centralize & Version Data

All images and annotations stored in one place with full version history

02
Annotate & Review

Built-in labeling with quality assurance workflows and team collaboration

03
Train & Compare

Run experiments with automatic logging of metrics, parameters, and artifacts

04
Deploy & Monitor

Ship models to production with full lineage and performance monitoring

04 — The Solution

How Picsellia delivered

Picsellia provided SGS with a unified platform that consolidated their entire computer vision workflow under one roof.

Unified Data Management

Datalake

All datasets, annotations, and metadata centralized in one platform with full version control. Every change is tracked, making experiments reproducible.

100%
Data traceability

Streamlined Annotation

Labeling Tool

Built-in labeling tools with quality assurance workflows. SGS's team can annotate, review, and validate data without leaving the platform.

Experiment Tracking

Experiment Tracking

Automatic logging of all training runs, hyperparameters, and metrics. Compare experiments side-by-side and never lose track of what worked.

0
Lost experiments

Team Collaboration

Dataset Management

Shared workspaces where all team members can access datasets, models, and experiments. Knowledge is preserved and accessible to everyone.

4→10+
Projects per team
05 — The Results

Business impact

The switch to Picsellia transformed SGS's computer vision operations, delivering measurable improvements across their entire workflow.

66% Faster Iterations

Model development cycles dropped from 3 days to 1 day. Faster iterations mean faster time-to-value for SGS clients.

Scalable Project Management

A 4-person team now effectively manages 10+ concurrent projects. Clear organization and shared workflows eliminate confusion.

Full Traceability

Every model can be traced back to its exact training data, parameters, and evaluation metrics. Critical for SGS's quality certification standards.

Knowledge Preservation

All experiments, decisions, and learnings are captured in the platform. New team members can onboard quickly and pick up where others left off.

Ready to achieve similar results?

See how Picsellia can help your team build and deploy computer vision models faster.