Dataloop Enhances Anomaly Detection with Flexible Data Pipelines | Case Study

logo black 1

About Dataloop

Dataloop is a technology company that builds data infrastructure and a data operating system for AI companies. Businesses of every size- from startups to public companies – use our cloud platform to accelerate the development and deployment of AI into production. Dataloop provides an end-to-end platform that covers the entire AI life cycle, from development to production. Our data-centric technology stack includes a data management, annotation platform that streamlines the process of generating data for deep learning and automation pipelines to accelerate AI projects to production, reducing costs and saving extensive engineering efforts on complex tools.

About is pioneering the Predictive Maintenance and Condition Based Monitoring markets with its visualization and AI platform. develops and manufactures customized visual solutions to organizations across a variety of industries in the form of highly resistant micro cameras and compute platforms with additional supplementary technologies. With the smallest cameras produced in the world, which are down to 1mm diameter including illumination, the high resolution technology has unique properties that have been authenticated by customers, such as NASA, in the strictest environmental conditions, including extreme temperatures, vibrations, and radiation. devices have been used across the medical, aerospace, industrial, research and defense industries. Founded in 2019, operates state-of-the-art production facilities.

Tackling the Anomaly Detection Process mainly deals with identifying, classifying and predicting behavioral and structural anomalies in complex mechanical systems. Their goal is to identify what these anomalies are and estimate their impact on the monitored system. 

With they’re focusing on two main aspects: verifying detection results and evaluating their proprietary models themselves. They are identifying what is missed, what is found, and what the thoughts of the model are on each anomaly detected. 

For example, rust detection can appear as a very stable part of a metal rod which is stationary in relation to the rod itself. However, rust can also appear as corrosion-dust, where the rust particles are spread along the rod regardless of their origin’s location. In this case, although the model’s confidence in the presence of rust is similar, a different classification between rust and corrosion-dust is required. has multiple models per use case. This is where Dataloop came in as a provider initially, in order to verify their model performance and provide measured KPI scores for their clients. Working closely with Dataloop, is able to detect critical issues saving millions of dollars in repairs and systems’ downtime. They are planning on introducing Dataloop as their integral data quality and model management engine, allowing them to improve life saving automations across critical industries such as aerospace, defense, railway and manufacturing.

Odysight Case Study

Avoiding a Tedious Manual Process with Dataloop’s Accelerated Data Pipelines large data volumes, which originate from multiple live sensors, require a high-throughput labeling pipelines, which may span anywhere from 10s to 1000s of images per hour per sensor. 

They  tested multiple tools and services from different annotation platforms and providers, but what made them choose Dataloop was the data pipelines and the flexibility of their pipelines feature. creates their own custom work environment on the Dataloop platform. Having a very well defined platform in Dataloop helped accelerate the configuration and deployment process itself, which was a real game changer for them.

How does Dataloop’s data pipelines work?

Dataloop's data pipelines are designed to streamline and accelerate the process of labeling and annotating large amounts of data. The pipelines are customizable and can be tailored to meet the specific needs of each client's data and use cases.

Step 1 | Data Ingestion: Ingest data from a variety of sources  Including image and video data, audio, & text files. Done via direct upload, cloud storage integration, or API integration.                                                          

Step 2 |  Data Preparation: Once data is ingested, it’s prepared for annotation. This may involve resizing images, converting file formats, or filtering out irrelevant data.                                                                         

 Step 3 | Auto-Annotation: Auto-annotations are considered at this point using the client model, this step will save time and money if done before sending it to a human annotator.                                                                                  

Step 4  | Data Annotation: The prepared data is then passed through Dataloop’s annotation tools, which allow human annotators to label and annotate the data according to the client’s specific requirements. The quality control phase can be improved using different methods such as smart sampling, real-time annotation validation, or post-processing validation. All methods save time and money using this automation integration into the workflow.      

Step 5 | Data Export: Once annotated, it is exported in the desired  format for use in machine learning models or other downstream applications. Dataloop’s pipelines can export data in a variety of formats, including JSON, CSV, and COCO.

It’s All About Saving Time and Money

 Once the recipe is defined and the task is known, it’s literally one step away from initiating the task because the dataset is always up-to-date. The initiation process took them a few days worth of setup. However, once completed, their once semi-manual process was now an automatic process. They simply launched it, and forgot about it. All the stress was taken off of their shoulders at this point.
Yeshiaya Odysight
“Working with Dataloop has proven to save Odysight.AI significantly in regards to money and time, by automating once tedious data-engineering and models’ evaluation procedures”
Yeshaia Zabaray, Senior Data Scientist at Odysight.AI

 This claim considers the time they saved with their data scientists including the data engineering aspect. Meaning, most of the manual work was taken care of and they didn’t need to use their DevOps team to push data in. Integration was also very quick which also saved us time and money.

“When it comes to our high throughput use cases (a few thousand images per day) Dataloop saved our data scientists 80-90% of their models evaluation time”, says Yishaia Zabary, Data Scientist,

They coordinated and ordered tasks via FAAS. For every new image, a FAAS calculated pre-annotation and a QA task was automatically created for that image and pre-annotation.

They plan on using pipelines in the future for image-augmentations in order to assist annotators. FAAS generates additional knowledge on the image (for example, for each pixel it adds a layer of texture score) to enrich the information given to annotators and ease and improve their annotations.

Easily Compare Your Model Performance

Dataloop’s comprehensive model management tool allows you to manage your model experiments and training, directly from where the data resides. By connecting your models to Dataloop you can then train them by using different versions of data, and evaluate model versions with test sets in order to compare your model performance. You can also easily deploy selected models and use them with pipelines for production or pre-annotation of additional items.
Odysight Case Study 1

How did use Dataloop’s pipelines and model management tools?

Trained a baseline model → for each new image the model generates appropriate annotations → passed to QA annotation task for evaluation and corrections → back to the model for retraining (based on the QA task results) → repeat.

They plan to use the new model management tool to keep track of some of their models’ versions. Given every improvement (using any of the procedures described above and more), the new version performance will be evaluated (using a dedicated test set and various metrics – also managed within the model management tool), and the most suitable version will be deployed to the appropriate FAAS/computational node(which is in our cloud base).
Odysight Case Study 2


Dataloop revolutionized's approach to anomaly detection and verification with their cutting-edge data infrastructure and operating system for AI companies. Dataloop's flexible data pipelines have proven to be a gamechanger for in their anomaly detection process. By providing an end-to-end platform that covers the entire lifecycle, Dataloop has enabled to accelerate their AI projects to production, reduce costs, and save extensive engineering efforts on complex tools. With Dataloop’s help, has been able to fix critical issues in complex machinery, saving millions of dollars in repairs. Additionally, Dataloop has saved drastically in regards to time and money every month. By automating their once-semi-manual process,'s data scientists have saved 80-90% of their model training and evaluation monthly time, and with the integration of pipelines, they plan to improve life-saving automations across critical industries. With Dataloop as their integral data quality and model management engine, plans to improve life-saving automations across critical industries such as aerospace, defense, railway, and manufacturing. Dataloop's advanced data infrastructure and operating system for AI companies have proven to be the missing piece of the puzzle for, enabling them to achieve breakthrough results and bring unparalleled value to their clients.

TFWNAH5TP U02JPPAAX7G a4eeba81db20 192

"Empowering businesses with a data-centric AI strategy lies at the core of our mission at Dataloop.

However, our impact goes beyond enabling AI production workflows; we're also dedicated to championing innovation and collaboration. This partnership with Yishaia and the team exemplifies the remarkable progress we can achieve when forward-thinking technology and a great team of innovative individuals unite.

As Dataloop's CSM, my focus is on creating customizable work environments in collaboration with our customers to drive their AI operations. With, we've already achieved significant milestones, and our journey is just beginning.

I'm very proud to be part of's efforts to address critical issues within complex machinery and the impact on how industries manage and maintain their machinery."

Adva Vivante, Customer Success Manager at Dataloop