Serverless AI-Powered Recommendation Engine with AWS Lambda and SageMaker
Joyanta Banerjee, Soumya Barman · International Journal of Computer Trends and Technology · 2024
The opportunities for e-commerce, entertainment, and many other services, which can be provided through the internet, have stimulated the growing need for recommendation systems to improve the experience of users. Such recommendation systems are based on complex matrices and need substantial equipment with high maintenance costs; therefore, they are hindered by scalability and performance constraints. Aims: We want to show how organizations could use serverless computing and leverage the exponential development of artificial intelligence to manage the scalability of effective recommendation systems with ease of deployment and usually low operating costs. Study Design: This paper describes the architecture and development of a serverless recommendation system for an e-commerce application based on AWS Lambda and SageMaker. The potential of the serverless to reduce costs, scale automatically, and be deployed and maintained easily is also investigated. Furthermore, we incorporate Amazon SageMaker for training, deploying, and managing machine learning models behind the recommendation engine. Place and Duration of Study: Organizations across various industries have implemented this approach in 2023 and 2024. Methodology: Collaborative, content-based filtering and the hybrid approach are employed in the recommendation process, and the results are generated in real-time. The complete application is built using the serverless computing model, in which AWS Lambda runs simple code in response to events or user interactions. In contrast, Amazon Sage Maker is used to train the models and make predictions. Exposing APIs is done with AWS API Gateway; storing users’ data is done with Amazon DynamoDB, while the model artifacts and the big data are stored in Amazon S3. Results: This architecture helps to avoid provisioning and managing servers, which makes the operation less complex. In this paper, we will describe all stages of the work, from data preprocessing to the generation of recommendations. Conclusion: The results thus demonstrate the exceptional scalability and responsiveness of the recommendation engine, capable of accommodating users’ real-time needs with trivial time delay.