How Dream Sports uses artificial intelligence
-> Our fan engagement product FanCode uses AI and ML backend framework to provide a personalised user experience to sports fans.
Founded in 2008, by Harsh Jain and Bhavit Sheth, Dream Sports is a sports technology company with brands such as Dream 11, FanCode, DreamX, and DreamSetGo.
“Across all our brands at Dream Sports, we implemented data-driven engineering with ML, AI and analytics very early in our automation journey, even though they were considered relatively new in the tech industry.”
Amit Sharma, CTO, Dream Sports
In a recent conversation with Analytics India Magazine(AIM), Amit Sharma, Chief Technology Officer of Dream Sports explained how the sports technology company is leveraging artificial intelligence(AI) and machine learning(ML) technologies to better its product and services delivery for users. An alumnus of Santa Clara University and the University of Massachusetts Dartmouth, Amit has earlier worked at Netflix and Yahoo.
AIM: How does your company drive growth by implementing AI/ML/Analytics?
Amit: At Dream Sports, one of our core culture pillars is being data-obsessed, and every decision we make is backed by data and technology. Last year, Dream11 won the Title Sponsorship for IPL 2020, and we saw the concurrency on Dream11 almost double to over 5.5 million, compared to the previous year. We also crossed the 110+ million users mark. To engage users at such a large scale, we used our data analytics platform to provide the best user experience by understanding their preferences, the features they like, etc.
The area where data has helped us in scaling last year is decision-making. For instance, as Dream Sports keeps growing, the number of decisions over various segments of users also grows multifold. In such situations, depending on humans for repeatable decisions is not a scalable solution. Instead, data-driven systems take care of such tasks efficiently. In addition, data analytics platforms can be leveraged for experimentation platforms, switching to algorithms, conversions, in-house data platform philosophy and product features (referral programs, social, product design, winner templates, etc.)
Our fan engagement product FanCode uses AI and ML backend framework to provide a personalised user experience to sports fans. It curates and shows content as per the preference of users, thereby making it easy to consume/engage sports data. Our payment solution – DreamPay uses AI and Big Data processing to make runtime decisions to increase the payment success rate for users.
AIM: Can you tell us about the composition of your team?
Amit: Across all our brands at Dream Sports, we implemented data-driven engineering with ML, AI and analytics very early in our automation journey, even though they were considered relatively new in the tech industry. Our AI/ML team currently consists of Applied Scientists, Machine Learning Engineers, and Decision Scientists. Multiple other teams contribute to building our platform in-house, such as the Data Engineers, Business Intelligence and the SDE (ML). This talent mix is approximately around 30% of our Sportan (our employees) Engineers.
We have recently hired more talent to our workforce and currently stand at about 700. However, we plan to hire over 200 more, of which about 80% will be hired for core roles in tech, product, and design.
AIM: What kind of skills do you look for while hiring data scientists?
Amit: Sports technology is a new, exciting and rapidly growing sector with a lot of opportunities. At Dream Sports, we look for talents who share a passion for sports, are data-driven, and are keen on building the latest cross-platform offerings.
We look for professionals who put users first and design innovative solutions. We are expanding our talent pool and look for professionals with experience in data warehousing, ML systems design, Big Data architecture, distributed systems, experimentation and Product Analytics, predictive algorithms, descriptive and diagnostic analytics, and research sciences like reinforcement learning, causal inference modelling.
AIM: What is the major challenge when it comes to finding talent?
AIM: The tech industry in India as well as globally is evolving fast, which has led to a scarcity of talent that is updated and at par with the latest skills in tech. While this gap could be filled with young tech professionals to a certain degree, experience in solving technical challenges has equal, if not more, weightage. Thus, having the right balance in terms of expertise and experience in addition to the keenness of learning and experimenting is highly sought after.
While many people have acquired talent around AI/ML/Data in the last 7-8 years, how to make such solutions work at scale remains a difficult task. We are facing a major lack of talent and ability to make AI-solutions work at scale. At Dream Sports, we look for people who strike the right balance between going deep and keeping up with the velocity at which our products evolve.
AIM: What does the toolstack of your team look like?
Amit: We actively use Python, Spark, R, Julia, Scala, PyTorch and TensorFlow. We have an in-house data platform on the Big Data ecosystem that includes acquiring, processing, storing and serving as data products or services.
- Data Highway is our major AWS Cloud-based in-house data acquisition framework for collecting transactional and clickstream data.
- We follow lambda architecture for real-time and batch use cases backed by Kafka as message bus, with Spark, Scala, Streaming and SQL Engine frameworks.
- Redshift is our integrated data warehouse and S3 as a data lake.
- Over time we have built multiple frameworks around data platforms such as Feature Store, Data Hub for the data dictionary, Data Aware for Funnel Analytics and in-house CDP as a data serving platform.
- We also use MLflow for model versioning and monitoring, and DataBricks is our integrated MLOps platform of choice.
- Python, Julia, R, Pytorch, Tensorflow, FastAI are extensively used for multiple use cases and graph databases for solving network-related problems.
AIM: What kind of challenges does your team face during the deployment stage?
Amit: The deployment of AI solutions at a large scale is a relatively new area of interest for us and the tech industry. The other challenge is to be cost-optimised. However, the combination of our in-house ML infrastructure, experimentation framework, data platform and data bricks have added immensely to address these challenges.
AIM: What is the significance of R&D at Dream Sports?
Amit: Over the last couple of years, R&D for backend, data and AI infrastructure has become an integral part of our workflow; we prefer to keep it in-house as data and tech are core to our business. As a data-driven team, our R&D team also helps us follow hyper-experimentation based on HEAL – Hypothesis, Experiment, Analysis and Learning. The approach enables us to fail fast, learn and build unique features for our users. We also remain open to working with highly evolved partners for new cutting-edge solutions that add great value to our brands.
AIM: How do you see the landscape of AI/ML evolving in India with regards to your domain?
Amit: Big Data and AI have already changed us all fundamentally. Every time we pick up the phone, there is some intelligent piece of code trying to suggest what we can do next with it, like when we should eat, what we should eat and even from where. Through experimentation, most tech companies always try to predictively find out what users want or what they are interested in doing.
Big Data and AI are also transforming technology by helping engineers automate things and create self-decision making and correcting systems. It is now time to focus not just on these changes but also on softer aspects of user touchpoints. Integrating empathy in user experiences, for instance, would be more important than anything else in the times we would be living in. While the focus is obviously on some fundamental challenges faced by AI models such as data distribution shifts, cause and effect etc., the softer side of things will play the most important part in the next set of transformations. The fairness of every algorithm that is used needs to be debated. We already live in an emotionally fractured society, but that doesn’t mean that data algorithms should reinforce the same. It is also time for professionals like us, with access to big data, to promote and emphasize more on the humanizing user experience through data.