Our vision at Index Exchange is total market efficiency, and our engineering teams are using machine learning and optimisation to design systems that pave the way for achieving just that.
I sat down with our team—Eran Udassin, VP, machine learning and optimisation engineering, Vance Wei, engineering lead manager, and Jaspreet Singh, senior data scientist—to learn more about the projects they’re working on, their approach to collaboration, and the powerful optimisations they’re bringing to our systems as well as the broader industry.
What’s a recent machine learning project that you found particularly rewarding?
Vance Wei: We’re building an optimised machine learning inference framework that can serve models at scale and with sub-millisecond latency, which is essential for tasks like personalised recommendations and real-time bidding in the ad tech ecosystem. The system handles heavy traffic smoothly, driving towards our vision of total market efficiency.
Working on this framework provides team members with the opportunity to work with cutting-edge technology, fostering valuable experience in managing high-speed, large-scale systems.
Jaspreet Singh: We’re using neural networks to scale our traffic shaping models, allowing us to learn from massive datasets. Using these advanced models helps us analyse smaller, more detailed segments of exchange traffic with greater accuracy, improving efficiency and performance across the board.
What tangible outcomes have our machine learning models made possible this year?
W: We moved our machine learning model training from the cloud to our own private cloud stack, successfully reducing our operational machine learning costs by more than 80%.
JS: We’re processing millions of unique ad requests per second with ultra-low latency, powered by our own data centers, models, and software. Our models are highly responsive to rapid market shifts and outages, giving us strong confidence we can handle traffic surges during streaming TV spikes and high-demand periods, such as the holiday season.
Our machine learning and optimisation team consists of Indexers from across the globe. How do you collaborate with each other and the rest of the business?
Eran Udassin: Our machine learning engineers and data scientists span across four countries and eight cities. We truly live and breathe global and distributed work. We collaborate with other teams by helping them optimise and reach their project goals, using existing and amended machine learning products.
JS: We plan tasks to the best of our ability so everyone can work within standard business hours for their location. Since we do span time zones, we sometimes use asynchronous communication when needed and will sync once a day when time zones overlap in working hours. We leverage a mixture of in-person meetings, Slack, and live documents to collaborate.
VW: We over-communicate in our team meetings. Things worth saying are worth saying twice. We’re also lucky to have a very transparent and trusting culture, which is tremendously helpful in bridging any physical distance between our teams.
Where do you see the greatest potential for machine learning to drive change at Index or across the ad tech industry?
EU: In the world of ad tech, buying and selling digital ad space involves many steps and players—marketers, media owners, and various intermediaries who facilitate the transaction. These intermediaries ensure that ads reach the right audiences, but they can add complexity and cost to the process.
By applying advanced machine learning, our engineering teams can simplify these transactions, much like how credit card systems work. When you swipe a credit card, the process is nearly instant: your bank verifies the payment, the card company processes it, and the vendor receives their funds—quickly and with minimal overhead. Machine learning can bring this same level of efficiency to ad transactions by automating decisions, reducing redundancy, and ensuring that each ad dollar is spent as effectively as possible.
Machine learning can also uncover patterns across the open internet, helping marketers tap into new opportunities. It can identify trends in how people browse, what content they engage with, or even niche audiences for specific products. By analysing these patterns, machine learning connects marketers with the right people, in the right place, making campaigns more effective and ads more relevant to consumers.
VW: Machine learning is already powering core efficiencies and decision-making at Index Exchange and across the ad tech industry. One promising area for industry-wide impact that remains largely unexplored is the interaction of machine learning products and algorithms across the various partners we work with, especially as reliance on machine learning for bidding and matching decisions grows.
For instance, if an exchange aims to anticipate users’ preferences while users simultaneously strive to understand the exchange offerings, this dynamic can create complex, multi-layered feedback loops shaped by each party’s objectives.
JS: A key area I personally see machine learning being at the center of is the ability to adapt and filter individual ad requests in real time as demand-side platforms (DSPs) change their campaigns and strategies. Being able to do this with a high level of precision will drive big efficiency wins for DSPs.
Overall, machine learning at this scale and velocity is rare and we have a huge opportunity to innovate not just for our customers but also for applied machine learning as a whole.
Learn more about the engineering teams at Index.
