Back to Blog

Meet Elie Raad, our Machine Learning Specialist!

WordLift is proud to have a team member of his caliber on board: let’s meet Elie💙

Quick Facts

Name: Elie Raad
Age: 38
Position in WordLift: Machine Learning Specialist
Languages spoken: Arabic, French, English, and just started learning the Italian language

Let’s Get to Know Elie

Tell us more about yourself…

Seeking new experiences and greeting new challenges that come my way has always been a big part of who I am. If you were to ask me, the journey began with my decision to study computer science, something that has fascinated me since a young age. An engineering degree and Ph.D. later, here I’m making the best of my professional life. So, on a typical day, you can find me sipping a cup of cold coffee and blending my CS background with my avid interest in R&D! Over the years, I’ve had many wonderful experiences professionally and met some amazing people from around the world. Many became good friends.

The opportunity to collaborate with folks hailing from multiple horizons and a myriad of cultures has been nothing but amazing. I could go on a limb here and say a part of me is a nomad- The spirit of freedom and the essence of independence is something I deeply value. Probably why I lived in different cities, and moved across three continents. I mean, what better way to authentically experience the local traditions & heritage! I’m based in France currently, yes bonjour! After all, it’s not too far from Italy, the headquarters of WordLift.

Tell us more about your professional interest…

I’ve been working in data science domains for many years and have developed an interest in the field of data and more specifically to synthetic data. I believe that through the rise of synthetic data, researchers and practitioners will unlock the true potential of artificial intelligence. In case this is something that piques your interest, I’ve written a blog on the past, present, and future of synthetic data.

How does your interest in synthetic data relate to your role at WordLift today?

AI-generated content is one form of synthetic data. Pre-trained language models have a wide range of application areas. One can use GPT-3 to produce AI-generated content, aka completions. When it comes to generating  high-quality completions that match the tone of voice of a brand, fine-tuning GPT-3 is the way to go. It’s a main part of my work at WordLift and this is how we proceed when generating completions for our clients. I recommend reading the in-depth blog on the topic of generating e-commerce product descriptions using GPT-3.

How do you see WordLift’s mission and what do you like about WordLift?

Building great products, combining the latest technologies, and working for something that makes a difference is my way of enjoying the process I’m part of. In line with my previous experiences where I built employment solutions to help people find jobs and then fintech solutions to get deserving people the credit they deserve. Now, with WordLift, I build AI-powered solutions to help entrepreneurs grow their businesses and brands reach a larger audience. To do so, WordLift’s products bring together the best of artificial intelligence, natural language processing, knowledge graphs and SEO. And this is changing the future of how to grow businesses and promote brands.

3 things you love the most about being a Wordlifter🏋️‍♀️

– Being part of a very nice and world-class team led by a visionary SEO, Andrea Volpini.

– Opportunity to work on interesting topics and the latest technologies. It’s a vibrant work environment.

– Ability to keep learning about different topics and technologies that, in some cases, go beyond the real work.

Can you share an exciting innovative work that you’re doing at WordLift?

Recently, I started exploring the power of DALL.E 2, OpenAI’s new AI system that can create realistic images from textual descriptions. Innovating and pushing things forward is part of WordLift’s culture. We run a new experiment to combine GPT-3 and DALL.E 2 in a single workflow. In a nutshell, we generate multiple completions using GPT-3 to describe one specific product. To pick the best one, all the completions are sent to DALL.E 2 to generate images. The last part of the workflow is to leverage visual similarity in order to pick the prompt, which refers to a GPT-3 completion, that enables DALL-E 2 to generate the most similar image. Jina AI’s CLIP-as-service is used to find the best visual match. In fact, the visual similarity is compared with the product’s original image. 

It’s a specific use case, yet in an early experimental phase, but it’s one interesting way to go beyond the seductive path of good enough when generating completions. Let’s take a moment to pause, imagine, and then appreciate the possibilities this can offer. 

To illustrate this in an entertaining way, here’s a comic that I generated using GPT-3 and DALL.E 2. The script is generated by GPT-3 while the illustrations are generated by DALL.E 2. 

Here’s a snapshot of the GPT-3 playground to generate an exchange between AI experts on the topic of synthetic data.

And here’s the prompt that I used to generate a first image sample of two AI experts having a discussion: “A comic illustration between two AI experts having a conversation. The bubbles are empty.”. DALL.E 2 generated a suggestion of six images. Afterwards, I triggered DALL.E 2 to generate multiple variations of one image from the suggestion set.

At the end, I used Canva to put together the AI text with the images in the comic format as shown below.

Something you want to share with us and our community…

One lesson I learnt and that I’d like to share is that less is more. It’s amazing the number of application domains where this principle holds true. Beyond its evident application for personal belongings, digital minimalism (use technology in an intentional way) and programming minimalism (keep the code clean and simple) are two other application domains. In real life, it’s not  always straightforward to identify what really matters. This requires an extra effort to reflect and reassess priorities. Without seeking such clarity it’s difficult to learn and distill knowledge in the age of information overload. As a knowledge worker, habits, deep work, and systems are the other important parts of my digital life to stay productive and keep learning.

Get in touch with Elie

Link to my Twitter / Personal website