Prior to co-founding Twiggle, Adi co-founded the networking infrastructure engineering group in Google Israel and led high impact search projects such as Spell Checker, Calculator, and Google Now. Today, Adi leads Twiggle’s technology strategy and heads a team of senior engineers focused on introducing new technologies and methodologies that constantly extend the value of Twiggle’s core search technology.
Adi: The technology cohort at Twiggle is working towards our grand mission: empowering retailers to deliver the most inspiring and delightful search experience to their customers. While our gifted R&D team is focusing on developing and improving our core knowledge-based search technology, the CTO team focuses on introducing new technologies and methodologies to our solutions. Specifically, we are working on three projects that make up our core differentiation and allow us to move the needle in e-commerce:
- Automated Ontology Building
Search that understands user intent across all eCommerce domains relies on both broad and deep knowledge. The power of Twiggle lies in the approach we take to representing knowledge in our unique ontological model. Building the foundation of modeling requires a close collaboration between our knowledge engineers and machines, but the rapid scaling of our understanding capabilities is done through automation and machine learning. Other solutions in the market rely on far shallower taxonomies that find the granularity of human intent challenging, especially in online shopping interactions.
- Automated mapping between catalogs and models
We knew early on that building a scalable product would require automated mapping between our customers’ catalogs and our knowledge model. That’s why we’re constantly working on improving and refining our automation process. Automation also applies to structuring the unstructured catalog data of our customers and allowing an efficient, seamless onboarding experience. Onboarding currently takes about 2-3 weeks, we are always looking for ways to make this shorter.
- Retail Semantic Spell Checker
Our Retail Semantic Spell Checker gives retailers two things they do not currently have: (1) Access to state-of-the-art spell checker and (2) output that corrects queries not only for spelling but also for meaning.
The numbers speak to the magnitude of the problem: 3-10% of queries have spelling mistakes. Spell checkers today can only fix an average of 50% of spelling errors. Our team is currently working on developing a spell correcting capability for retailers that can semantically spell-correct 70% of mistakes. Our algorithms are designed to process most of the queries within less than a millisecond so impact on latency is extremely minimal.
Our goal is to build a semantic spell checker tailored to each customer, taking into account collections and product lines unique to that retailer.
What do you think is the biggest misconception about search?
Adi: The biggest misconception is that Google solved search and there is no room for further innovation. However, wanting to know if it’s raining in China is very different from wanting to buy a raincoat. The reality is that eCommerce search has a lot of unfulfilled potential. In a world where expectations are growing for machines to fully understand humans, eCommerce lags.
Unlike search for facts (e.g. what is the weather in Tel Aviv right now?), search for products is a psychological journey. As such, the words we use to describe what we are looking for has meaning requiring analysis. It can’t just be about keywords anymore. Search should be based on relationships and associations. This is the kind of search we build at Twiggle.
Why is e-commerce search particularly difficult?
Adi: e-Commerce search is surprisingly complicated and multilayered. If I’m trying to get movie tickets and look up showtimes, Google pulls up an accurate list of times from nearby theaters. But if I type into a retail search engine, “I’m looking for a blue dress with sleeves in size small for a wedding,” I’ll get plenty of irrelevant results, or even worse – nothing at all. e-Commerce search today does not behave like a store clerk that understands what I’m looking for and get me to the closest option available in the store. Imagine what consumer experience could look like if search engines could understand the way store clerks do.
Why did Twiggle decide to focus on e-commerce as a vertical, as opposed to creating a horizontal search solution like Google?
Adi: We believe that the best search experience requires “knowledge based” technology. Since the human knowledge domain is so huge, almost unfathomable, we decided to break it into small pieces and start where the vacuum in understanding is biggest – in the eCommerce domain. We let Google be Google — our goal is to do great things within the retail space. It is a highly ambitious goal but we are already seeing the impact of our technology on our customers’ performance and ability to compete effectively.
How is Twiggle able to hire such great people?
Adi: The best people want to work on the hardest problems in a tech-first environment, and this is what we offer.
We’re able to attract the top engineers and data scientists to Twiggle by appealing to their need to work on the most complex problems in NLP and deep learning. We encourage our people to establish their authority in their topics of interest and expertise. We provide them with the opportunity to work with other brilliant people who excel in what they do and from whom they can learn in a meritocratic environment of coaching and caring. Our people know that they have the chance to make a real impact on an entire industry. Because of the complexity and scope of our challenge, each person in the company is critical to our success — and they can feel it.
What excites you most about the future of the company?
Adi: Envisioning where we can go with such powerful technology. The possibilities are truly endless.
Where do you see Twiggle in five years?
Adi: In much bigger offices.