See what we are working on below.

PADL is a unifying open-source development framework for PyTorch for building, training, serializing, and applying deep learning models. It streamlines the entire deep learning workflow, from experimentation to deployment.

The library comprises a functional API, and compact primitives for composing pre-processing steps, model passes, and post-processing steps into a single formalism. This leads to a very handy separation of concerns between development and production code, as well as reflecting an intuitive mental model; the macrostructure of a model is conceived of as graph-based and functional, whereas the fine details are conceived of as procedural.

In addition, PADL models come with all steps (pre-processing, model layers, and post-processing) necessary for reproducibility baked into the model objects. This leads to fewer errors from mismatches between models and pre- and post-processing steps, in addition to fewer versioning errors.

Project website:

Aleph Search is a fully AI-based product search and discovery suite for e-commerce. The services include text search, reverse-image search, similar product recommendation, product tagging among others.

The core of Aleph Search is our proprietary “product fingerprint” algorithm which builds a representation of e-commerce products using the same set of faculties that human users use to understand products: through analysis/ perception of a product’s images, attributes, categorizations, and textual descriptions.

Aleph search builds on the latest discoveries in deep learning for natural language processing and computer vision and incorporates ensembles of neural networks tailor-made for the e-commerce industry and improving user experience.

After two years of research and development, our Aleph Search IP was acquired by the British Attraqt Group PLC, a leading player in this field. It is now used by several of the top 20 online shops in Europe allowing millions of users to daily benefit from an unparalleled shopping experience.

Using the latest AI architectures like GANs and variational autoencoders we want to enable artists to create visual art in novel ways.

We are working on a system which understands the essence of any content enabling the intelligent ingestion and processing of vast amounts of information from versatile sources.