July 2025

Features 🚀:

AI Review is Live!

You can now precisely evaluate your model’s performance in production, task by task. The goal: pinpoint your AI’s weaknesses, improve data quality, and continuously boost performance.

🎯 Why AI Review?

Because you can't improve what you don't measure. AI Review lets you:

  • See exactly how AI results appear in real production workflows

  • Review and correct tasks directly from the interface

  • Build a reliable ground truth

  • Track performance by error type, task, or over time

？ How does it work?

  1. Set up your review campaigns: choose which tasks to evaluate

  2. Run your reviews: validate or correct AI outputs

  3. Explore the results: analyse performance per task

👇 See it in action

AI Review is the first milestone of AI Boost, our global initiative aimed at enhancing AI performance in production. Now that performance can be measured, the next step is to provide access to reviewed images and enable integration of the feedback loop for continuous model improvement.

Audit — Spot and fix annotation errors

🎯 Why it matters

Training high-performing models starts with proper data. Even small annotation errors can hurt performance. The Audit feature helps you automatically find and fix mistakes in your dataset — faster and more reliably.

It brings immediate value:

  • ⏱️ Saves time for annotation teams by surfacing only the most likely issues

  • 📈 Boosts model performance by ensuring better training data

？How does it work?

  1. Go to the Audit tab

  2. Choose a trained model

  3. The platform compares predictions to your annotations

  4. It flags potential errors:

    • Missing annotations: predicted but not annotated

    • Extra annotations: annotated but not predicted

Then, review each image, accept or reject corrections — and your dataset gets better in minutes.

👇 See it in action :

The SFTP data export

🎯 Why it matters

Whether for business intelligence, supporting your processes, or simply for archiving purposes, the data export allows you to get all your Deepomatic data :

  • at regular intervals (daily, weekly, monthly)

  • in a structured format (jsoline and jpeg in dedicated folders)

  • to the destination of your choice (SFTP server, AWS S3, GCP bucket, Azure BLOB storage.

? How does it works ?

  1. Your organization provides Deepomatic the details of your SFTP server.

  2. Deepomatic sets up the data export according to your needs.

  3. You receive your data at the chosen interval in a BI friendly format.

See all the documentation here : Data export

Last updated

Was this helpful?