The Coveo Relevance Cloud™ is a market-leading AI-powered relevance platform. We aim to enable our customers to deliver the relevant experiences that we believe people expect in the new digital economy. Our cloud-native, multi-tenant SaaS platform injects search, recommendations, and personalization solutions into digital experiences. We provide solutions for ecommerce, service, website, and workplace applications. These are designed to provide tangible value to our customers by helping drive revenue growth, reduce customer support costs, increase customer satisfaction and website engagement, and improve employee proficiency and satisfaction.
Coveo is an AI as a service search engine -- improving the digital experience that our client companies provide to help them remain competitive to major websites. Without extensive pipelines and larger datasets, we want to maximize effectiveness of search and recommendation systems given their set of unique constraints and strategies.
Coveo’s main business can be segmented into two distinct streams. Our service component allows people to search for solutions easily on our client’s website, and also equip support agents with contextual information to solve a particular case quickly. In our commerce component, we aim to level the playing field for companies who cannot afford developing expensive algorithms in-house and are undergoing transformation, yet still wish to serve users well, by providing tools that enable high-quality personalized search and interactions.
In the last three years, NLP has been shifted entirely by new language models, and it’s the perfect time for Coveo to bring these types of models into production whilst maintaining fast queries times (<15ms) and covering the variety of use-cases in many languages. In addition, because of the wide ranging scale of our clients, we are interested in fine-tuning on smaller datasets and knowledge transfer generally in our machine learning teams.
Although commerce is booming overall, searches for many sites are quite poor, and most users only visit once. As a result, companies miss cross-selling opportunities and entry-points to generate recurring revenue. The lack of resources to optimize models and infrastructure to support user experience is especially detrimental to smaller players in the market. That’s why Coveo customizes the experience live for new users to the website by leveraging session data to create a better multi-dimensional picture of the user, then apply machine learning on the data space, and finally power downstream tasks such as reranking results or serving relevant recommendations. Join Coveo to empower the next generation of digital commerce.
Personalization as you go is one area where Coveo is creative and innovative. Our objective is not always to publish something; rather we focus on how to solve a particular problem. We like to be transparent and validate the ideas in our community and brainstorm before getting started. For example, we worked on personalizing the visit of users we only see once by first finding a way to learn their intentions. In this case, innovation can take the form of new algorithms, or even just applying existing algorithms well in a novel angle, all under production constraints of speed and unknown variables.
Our team of product managers are always in contact with clients and monitor the market closely. Together with the R&D management team, major feature additions or changes are decided. If the task is feasible, “how” and execution that follows is determined and completed by the ML team. Apart from this top-down approach, another kind of product improvement at Coveo is bottom-up. Production teams drive this by conducting reviews of new technologies, and subsequently discover opportunities for the technologies’ application in their respective market and line of business, which they surface up to managers for approval. If it makes sense, it is structured into a project and pursued.
Since ML models need many surrounding components to function, we can have the ML PM, API PM, and frontend PM, all working synchronously on one project. One recent example is an algorithm that selects the correct category for a query. The ML team is needed to design the decision-making process; the user analytics team is needed to track feedback; the API team is needed to answer the call; the UI team is needed to develop the component that adjusts automatically for category selection. The entire project is cross-functionally managed by the product team.
We give freedom to anyone on our engineering teams to try out new tools. When we see new best practices, we are open to do internal research projects and proof of concepts. There are always code-reviews, in which at least two people need to validate each line of code before deploying to production. Code review is not only a time to standardize, but also a great vehicle to share knowledge between and within teams. This is one of the main ways our interns learn about the inner-workings of Coveo--being assigned small features and walking through code reviews.
We like to grow the people within the company, so most promotions in the past have been internal. Coveo follows a career growth framework, where we define expectations of different levels, and there is guidance for progress and things employees should learn or do to level up. It is common to have people graduating from interns to developers, senior developers, to team lead or staff developer. However, at one point you must decide to either stay purely technical, or start to manage teams and people; we love both types of progressions. We also like to hire senior people that bring different perspectives or possess a particular skill set from time to time. But if there is someone internally who can do the job, we prefer promotion.
Almost half the company at Coveo (around 250-300) is on the R&D team. Even in a pandemic, we invested and doubled-down on our product development, due to the conviction that sales will increase with an exceptional and stronger product. There is an emphasis in R&D because we want to be the best in our field. And since software is winner takes all, Coveo is building a robust R&D culture to keep up with the rapid technological change.
Transparency is very important at Coveo. Some practices to help us establish high bandwidth communication are daily scrum meetings and random-coffee chats at the company level. In order to create a safe environment, we designed a flat hierarchy where individuals can always speak to their director or VP, and always be listened to.
We encourage everyone to pitch their ideas that can turn into projects and features. We also want people to speak up in design reviews (there is a good way to bring up even bad ideas), stand up for their opinions backed by rationales, and share openingly in general. Constructive disagreements and arguments (that are respectful and productive) often occur, and we feel comfortable with them. Candidates will experience this first-hand during our interview and evaluation process when we ask them questions regarding their take-home solution.
Coveo’s feedback processes include:
Lastly, there are quarterly team activities, such as dinner before the pandemic, and online games now. We want the team to get to know each other outside of jobs and foster meaningful relationships around topics that are unrelated to work.
Internships: Recruitment team completes a phone screen with the candidate and reviews their CV. If they are a good fit, an 1h will be set up with a general R&D interviewer and an individual from a specific team that they may fit. A decision is then made.
Full-time: Similar to internships, there is a recruiter phone screen followed by a first interview with R&D interviewers (developers within our teams) to evaluate cultural fit, main interests of the candidate, and dive into their experiences. Afterwards, the candidate will receive a technical test (front-end, back-end, standard ML, or NLP). For ML and NLP, this is a take-home with a dataset and small problem. We are looking to see how they analyze the problem, tease out patterns, and programming abilities. Because we don’t expect you to have the full solution, we also like to ask future approaches you might try and what would you do differently next time. The next interview is a presentation given by the candidate on their work and the chance for our team to ask questions. If they pass this stage, there is a final conversation with the R&D VP, in which placement and compensation can be discussed.
In summary, 1. Recruiter phone screen 2. R&D behaviour and conceptual interview 3. Technical assessment 4. Technical assessment presentation 5. Final interview.
Our ML teams are divided according to use-cases and we try to keep them small (an average of 5-6/team plus or minus 2/3). Each team is responsible for experimentation, production inference, and iterating from customer feedback, while all collaborating with the ML platform team who owns infrastructure and serving. Use-cases form different teams include Question & Answering, Recommendations, Search (Rankings).
https://blog.coveo.com/from-research-to-production-how-our-ai-helps-businesses/
