Category hierarchy

Click-optimizing hierarchical category prediction

E-commerce or classified sites often contain large category hierarchies of their goods. When a user wants to submit a new item, our AI-based system can suggest up to N categories (usually 1 to 5). Smart category prediction can speed up this process by 70%, saving users precious time.

The base of this technology is a machine learning system that predicts the leaf node category based on a combination of image and/or text. Our proprietary algorithm then chooses optimal nodes from the hierarchy tree to minimize the number of user’s clicks through the hierarchy to get to the correct category.

Image Search

Image Search (or Visual Search) is an AI (Artificial Intelligence) powered technology that enables users to sift through millions of items in order to find the best visually matching items. After the user gives a query image, the search process takes a few fractions of a second and delivers visually and semantically relevant results.

Velebit AI offers off-the-shelf models for general categories and different niches like fashion and furniture. But the best results are achieved by fine-tuning our models to your domain data.

Some of the Image Search use cases include:

  • Visual recommender
  • Visual browsing
  • Image and Text-based similarity search
  • Style suggestions
  • Camera Search
    • Fashion Cam, Furniture Cam
  • Personalized Newsletter
    • Based on visually similar items
Image Search Example
Integration with databases

Integration of Machine Learning with your databases

Velebit AI offers custom AI-facilitating plugins for ElasticSearch, Solr, Postgres and MySQL. This way you can seamlessly integrate our text or vision-based services to your existing searches in your DB. No need to migrate your data, or use a new system. Our plugins will deliver fast and efficient queries on the predictions of our AI systems.

Recommender Systems

Besides traditional recommender systems that use collaborative filtering to recommend items that users frequently bought together, Velebit AI offers item-to-item recommenders that use images and text processed with deep learning to recommend most similar items (based on visual similarity, text, or combination of both).

Recommender systems icon

Real-time object detection and tracking

Whether you build a professional application for managing free car parking spaces or a consumer mobile app for learning a new language, object detection, and tracking system can raise it to a new level. Object detection systems can replace expensive sensors or provide new information that was previously non-measurable. For consumer apps, it delivers a much more engaging user experience.

On the left you can see an example of our Visual Translator Mobile Application - Babbly.

Tag suggestions

Many of today's CMS systems are using tags to describe, sort and categorize articles or inventory items. Finding appropriate tags can be a tedious job, and tag quality can often suffer because of it, reducing the SEO and marketing potential of a site.

Our solution suggests tags based on text, image, or combination of both to optimize different metrics. It can also be used to manage the existing tag database by enabling faster tag merging and removal, thus improving tag quality.

Example of an article

A new era of supersonic flight might be just around the corner, but there are three challenges to overcome when it comes to flying faster than the speed of sound.

Those are the three Es of aviation: engineering, environment and economics.

Concorde, the aeronautical marvel that made its last flight 16 years ago this week, only conquered the first of those three travel challenges.

The world's slinkiest airliner could transport passengers across the Atlantic in less than half the time taken by other commercial aircraft, but it still had ecological shortcomings and high operating costs.


Predicted tags

  • plane
  • aeronautics
  • Concorde
  • supersonic-travel
  • jet
Weather clouds

Predictive modelling of dynamic systems

Machine Learning and especially Deep Learning can improve and speed up the simulations of linear and non-linear dynamic systems. Their advantages over classical numerical methods are:

  • 5x-50x faster compute time
  • accurate at short timescales
  • reduced hardware requirements

Predicting the weather or simulating fluids are just some of the tasks where our models beat classical methods. Deep learning algorithms often used here are 3D convolutional neural networks and generative adversarial networks (GANs).

Data Science and Machine Learning consultancies

Planning a new strategy for your company/department? Designing a new product or redesigning an existing one? Our experts can help you to extract more value from your product or service.

Creating and maintaining a dedicated machine learning or data science department incurs large time and expenditures costs, especially for companies that possess little knowledge in this field. In such environments, workshops with a team of experts can provide a lot of value and open some new horizons on how to best take advantage of innovative technologies.

Image of Velebit AI expert

We build AI for your needs.

Meet our highly experienced team who just loves to build AI and design its surrounding to incorporate it in your business. Find out for your self how much you can benefit from our fair and open approach.

Contact Us