March 21, 2022

Investment notes: Deci US$9.1m Seed

We have co-led Deci’s USD $9.1m Seed round. We are excited to be joining Yonatan, Joe, Ran and the team

Philippe Schwartz

October 28, 2020

Investment notes: Deci US$9.1m Seed

We have co-led Deci’s USD $9.1m Seed round. We are excited to be joining Yonatan, Joe, Ran and the team

Philippe Schwartz

Advancements in AI, powered by deep learning, have triggered ground-breaking innovations. But, long development cycles, high compute costs, and poor inference performance are making it difficult for enterprises to productise AI. That was until the team at Deci realised that the solution to breaking the AI barrier, lay in harnessing the AI itself.

We have co-led Deci’s USD $9.1m Seed round. We are excited to be joining Yonatan, Joe, Ran and the team.

The problem

We are living through one of the greatest advances in technology: the advent of artificial intelligence.

The forces driving this intense progress are the deep neural networks (DNNs) and powerful computing that enable superior modelling and prediction. These new deep learning models are enabling unprecedented machine vision capabilities, such as accurate perception functionalities essential for many low-level visual recognition tasks such as object classification, detection and tracking, and even the recognition of visual relationships in images and videos. Developments in performance in these tasks have enabled powerful applications such as autonomous driving, video analytics, security control and monitoring, smart home and city, automated shopping (e.g., Amazon Go), medical diagnostics, conversational AI applications, and more.

The effective deployment and operation of DNNs in commercial applications depend on high performance, in terms of both accuracy and efficiency.

This is usually where the problems begin.

Traditional machine learning model development is resource-intensive, requiring significant domain knowledge and time to produce and compare dozens of models. A simple data point that illustrates this pain: close to 90% of deep learning models do not make it to production either because the costs to serve these models are prohibitive, the company doesnʼt have the skills, or the performance post-deployment is not good enough.

And yet, more organisations are finding ML algorithms improve their product or service, e.g. Zillow performing image analysis to improve the quality of their listing or JP Morgan building a model for credit risk analysis.

And so the question becomes, how do companies create high-quality neural networks that are affordable in computing costs and simple to develop and maintain?

We believe we found the answer with the team at Deci.

What Deci does

Deci uses AI to supercharge AI models.

The team have built an optimisation engine called AutoNAC (or Automated Neural Architecture Construction) that optimises deep learning models to more effectively use their hardware platform, be it CPU, GPU, or special purpose ASIC accelerators.

It takes as input a user-trained deep neural network, a dataset, and access to an inference platform. It then redesigns the user’s neural network to derive an optimised architecture whose latency is substantially lower, without compromising accuracy and enabling models to reach their performance sweet spot for any hardware and data organisations are using.

The beauty of the Deci solution is that it is also generic: it works on any hardware and with any type of data source and is therefore adapted to most attractive AI fields such as image processing, NLP, time series, etc.

Let’s use a Deci customer example to illustrate the power of Deci’s solution.

WSC Sports is an AI-driven platform that turns live sports events into short videos for iconic organisations such as NBA, Cricket Australia, ESPN and more. At the core of WSC’s platform is a data science team who build complex deep learning models to help select and package the best moments from every game. The company was seeing high costs for one of their novel deep learning applications in the cloud, due to the large number of GPUs needed for production. They used Deci’s AutoNAC optimisation engine to transform the efficiency and speed of their deep learning models and saved 78% in Cloud Computing costs. That is huge.

The team

Behind Deci is a truly extraordinary team. When we first met Yonatan, Deci was an idea. But in just a few short months, the team transformed their white papers into a business, signing up significant customers such as Intel and Lightricks and hiring an impressive team to build out their capabilities.

Yonatan Geifman is the co-founder and CEO at Deci. He holds three degrees and a PhD. Before Deci, he designed and taught a deep-learning course with co-founder Prof. Ran El-Yaniv at Technion, the Israel Institute of Technology, and a successful career in the Israeli Defence Force.

Jonathan (Joe) Elial is the co-founder and COO at Deci. Before Deci, he served for 11 years in the Israeli Air Force, leaving as the Head of Operations and in senior product and R&D roles at Mercedes Benz and Axon Vision.

Ran El-Yaniv is the co-founder and Chief Scientist at Deci. He remains a Professor of Computer Science at Technion University, where he has been teaching for 22 years. Before Deci, he was also a Visiting Staff Research Scientist at Google.

They are uniquely qualified to go after their vision, and we’re so thrilled to be on the journey.

Find out more, at