top of page

Meet Gilad Barkan: An AI-Driven Interview with Wix’s Head of Data Science Guild



As a followup to our meetup “AI @ Wix: A Cross Review of Data Science, Engineering and Generative AI at Wix”, we sat with Gilad Barkan who leads Wix’s Data Science Guild, for an insightful AI-driven interview. We discussed Data Science initiatives, the future of AI, and important considerations such as impact, talent, and collaboration.


Gilad Barkan Wix Engineering


Hi Gilad, tell us a bit about yourself and at a high level about the Data Science Guild you’re leading.


Hi, I am Gilad, a machine learning practitioner and leader with more than two decades of experience in the industry and academy. Data Science at Wix is split into two main components: the guild and the Data Science group. The guild consists of data scientists and AI researchers, while the Data Science group includes not only data scientists but also DataOps and MLOps professionals. The guild is responsible for its members' professionalism, growth and standards, the group develops projects in different areas within Wix.



Can you share about the current state of AI at Wix and what Wix is currently working on to leverage Generative AI?


We're in the beginning of a revolution caused by Generative AI (aka GenAI) due to the appearance of ChatGPT. It's a revolution because it's the first time AI is accessible not only to data scientists but to basically everyone, through APIs. For the first time, AI is democratized and available for everyone as a true and valuable service.


However, it's important to clarify that while many people are just now being exposed to it, AI has already been around for quite some time, including at Wix. Wix's first AI based product - ADI (Artificial Design Intelligence) - was launched back in 2016. Since then, Wix has built a top class AI organization, consisting of dozens of data scientists and AI researchers. We have established the necessary infrastructure, both engineering-wise (aka MLOps) and data-wise (aka DataOps), making it possible to already deploy to production 200+ AI models in more than 20 verticals.


AI is practically instanced by models that capture intelligence. Previously, these models could only be developed by the data scientists of a company. But now, theoretically, with the advent of the so-called Foundation Models, in general, and LLMs (Large Language Models), in particular, any engineer can directly access these models via APIs and very quickly develop AI based applications. This is a huge enabler for a company like Wix and I'll explain why.


Wix is an online platform for building an online presence for any type of site, for any type of user. From Self Creators building their portfolios, to individuals running small businesses and services, to professionals building advanced websites with customized code.


Wix is an all-in-one platform that encapsulates dozens of sub products that users interact with. It could be adding products to their store, placing a booking service, adding content like textual product descriptions, polished marketing materials, personalized images, etc. All of these functionalities take time to create and can sometimes be challenging for our users.


Among the models we've already deployed we have full site creation and site design, NLP based projects like text creation, user support, user sentiment, site classification, computer vision based products like image manipulation and auto-enhancement, recommendation engines, semantic search, time-series forecasting, and some strategic initiatives that I can not share at the moment.


As early adopters of AI technologies, we've collaborated in the past on joint research efforts with major companies like GoogleX & IBM, as well as have already released GPT-based features in production already at the beginning of this year.


Data Science Wix Engineering

From your perspective, how do you envision AI transforming the industry in the next five to ten years, and what steps are we taking to stay ahead of these changes?


We see the following four main pillars where GenAI would bring value and uplift our business the most:

  1. Helping our users to overcome one of their most significant barriers - generating content for their websites.

  2. Changing the way users interact with our products and services. First, shifting from understanding computer systems to natural language interfaces; Second, moving from DIY to AI assistant that Do It For You (DIFY), using emerging technologies like agents and tools to make this possible.

  3. Website creation and design. As a website platform, our product involves multi modalities like text, images, design and layouts. We cannot currently share much about our efforts here but we definitely see the GenAI technologies progressively merging these modalities together more natively.

  4. Accelerating product development by leveraging productivity coding tools like GitHub CoPilot, enabling us to do the same things we did previously, but now much faster, on a bigger scale and with improved performance.


GenAI is shaking the ground nowadays and truly gives new functionalities to basically everyone. However, if companies didn't truly invest in AI before this dramatic point in time, then they'll soon face the fact that it will take them some time to execute a true change using AI within their products. We believe we'll reap the benefits of GenAI and be a leader in our industry because we've already planted the AI seeds many years ago.


We've built a state-of-the-art organization of data scientists and AI researchers, along with a dedicated Machine Learning engineering Platform we've built in-house, as well as a data-centric organization. Data-centric AI was a big hype recently until GenAI came. However, five years ago, we recognized this approach to AI and built a data-centric organization with dedicated teams for data collection (aka DataOps). You can read and listen more about it here in our podcast:



Trying to envision what AI will look like several years ahead is very challenging. We're only at the beginning of this new revolution. Look what happened in just six months! We’re starting to see some sparks of true AGI (Artificial General Intelligence) within GPT-4 and Chain-of-Thoughts like reasoning. Once we reach the point of AGI with true human-like reasoning capabilities, everything will change. Not only our industry will be changed but all society and humanity is going to be completely disrupted. I'll use Yogi Bera's quote: "It is difficult to make predictions, especially about the future" and add to it: "nowadays more than ever, even only within the horizon of couple of years ahead."



What are the key challenges you see that come with Generative AI, both in terms of engineering and technological advancements?


GenAI is a revolutionary and disruptive technology because it shifts the paradigms of doing the same things done before with traditional AI. This technology imitates some human capabilities very well and can potentially dramatically speed up the development of new products. For companies to adopt this technology as fast as possible we observed the following gaps which need to be mitigated:


  • Gap in infrastructure. There is a need for a GenAI specific engineering infrastructure, which we term LLMOps. We're already working on extending our current MLOps platform to accommodate all the new GenAI functionalities, such as prompts management, LLM provider agnostic wrappers, evaluation suite, etc.

  • The 60-5 gap between a (extremely) fast PoC and a production ready product. The 60-5 gap is the GenAI version of the known 80-20 Pareto rule. We already see how incredibly easy and fast it is for engineers to build a ~60% performance PoC of an AI based application using AI-as-a-Service (external API calls to LLMs). However, we observe as well how hard it is to make the extra 40% to make it production ready. Black box LLMs present challenges such as the possibility for them to hallucinate, it may be hard to control them, some of them may not be updated, it may be hard to evaluate them before going to production, they're not steady state machines (i.e. a small change in prompt may lead to totally different results), and more. We observe in general the hype around, where developers think they can now build a startup / solutions based on AI, but we observe how they get stuck since they don’t have the practice of building AI applications.

  • Gap in knowledge. The AI dev cycle is different from the regular engineering one. It's not deterministic, it can involve lengthy and iteratively research cycles, etc. To create an impact of AI at Wix along the years, we continuously invested in educating the organization around AI. We gave lectures to engineers, we created a course for business analysts, and we educated PMs. However, since everyone now can theoretically develop AI based products by themselves, there is a gap in AI knowledge to be filled throughout the whole organization, to eventually succeed running a project without a data scientist in the loop. For example, what is the AI product life cycle, how to evaluate, best practices for prompt engineering, etc. We've built an onboarding educational kit namely: 'What should I know to run an AI project?' This kit includes the concepts, best practices, the technical details, references to online material, etc.


What is the role of Data Science in the era of Generative AI?


We split the GenAI landscape between AI Users and AI Creators. AI users are 95% of the company - non data scientists who can now, for the first time, build AI-based products by calling external APIs and using prompt engineering to accomplish their tasks. AI Creators are 5% of the company - data scientists that create value for the company in the places that AI users can not. AI creators have two primary roles in this new era:


- First is to be the enablers for AI Users, which means:

  • Building platformized versions of common use cases, such as Q&A service on top of internal documentation and knowledge-base databases.

  • Educating the AI users so they can fill the '60-5' gap by themselves without data scientist in-the-loop.

  • We mark this responsibility as critical for accelerating the release of AI based products at scale.


- Second is for AI creators to focus on creating value for the organization and serve as the AI experts who keep themselves up-to-date with the current exponential developments; Within the territories dedicated to AI Value Creators, we observe the following areas:

  • Building models, mostly M(edium)LMs and fine tuning LLMs;

  • Engaging in research to discover the company's next AI-based ground breaking technologies;

  • Developing methods and techniques for better controllability and reliability of the LLMs;



How do you collaborate with other guilds or teams within Wix to drive AI-driven initiatives, and how important is cross-functional collaboration in achieving our data science goals?


As the AI experts in the company, we are the native mentors and consultants for any AI project / question. To run fast and efficiently, we leverage the power of Wix as a big tech-savvy company. Wix's matrix structure of guilds (professions) and companies (verticals) allows the companies to run fast and be focused on their targets without being dependent on other optional blockers.


However, the option for everyone to run AI projects now makes it possible for the creation of lots of AI products, inevitably many of them are basically the same type of projects.


To successfully run fast and keep the whole company up-to-date, we've established a group of AI ambassadors - representatives from each vertical at Wix - whom we educate and touch base with frequently to sync. We found this way of distributing the knowledge already successful in other previous horizontal projects at Wix.



From your experience how any developer should get ready for what the future holds for us?


We believe that AI technologies should be part of the toolbox of every developer today. We see it more and more how AI is becoming like a personal assistant to allow us to do the same stuff we did before, but in a much faster and more accurate way. Using tools like GitHub CoPilot are disruptive to the way a developer works and are revolutionizing the development process. Therefore, developers, on the one hand, need to keep up-to-date with the AI tools out there. While on the other hand, acquiring a deeper understanding of AI technologies in order to develop breakthrough technologies using AI.



 

More of Wix Engineering's updates and insights:


Yorumlar


bottom of page