During the Nvidia Summit in Mumbai, Meesho, a leading eCommerce company, made a significant announcement. Debdoot Mukherjee, the head of AI at Meesho, revealed that the company will be open-sourcing its ML platform, including its feature store and model orchestrator. This move is expected to empower developers and data scientists worldwide to utilize AI more effectively. In a presentation titled “Cost Effective High-Performance Model Serving for the Masses,” Debdoot delved into Meesho’s “massive scale” machine learning models. He discussed the Feature Store, an architecture designed for high-performance ML operations, which helps Meesho standardize and manage features for their ML models at scale. The presentation also highlighted that this system can handle millions of requests per second, with 99% of them being processed in under 10 milliseconds. Debdoot also announced that the Feature Store architecture will be open-sourced. In a LinkedIn post, he mentioned that they focused on scaling transformer-based ranking models in production and optimizing them with TensorRT to meet strict latency and cost requirements. Recently, at the Build With Meta Summit, Meta revealed that Meesho is one of the many consumer apps planning to implement Meta’s open-source Llama models in their tech stack. Over the years, Meesho has actively integrated generative AI and ML capabilities into its platform. Their large-scale decision engine has incorporated various features to improve code maintainability, reduce system latency, and provide an enhanced error-handling system with multiple fallback solutions. Additionally, the decision engine helps developers monitor system health by creating a dashboard to track and identify issues easily. On the user front, Meesho uses generative AI to simplify the listing process for sellers on the platform. They have also implemented features to translate product descriptions, addresses, and voice searches into other Indian languages besides Hindi. Meesho also leverages Microsoft’s Azure OpenAI to optimize call center costs and operations, resulting in a reported 20% increase in customer effort scores. Last year, Meesho collaborated with IISc to advance their research in generative AI and develop multimodal representation learning and other generative AI capabilities to enhance the user experience. In an interview with AIM, Debdoot stated that AI and ML are integral components in solving a wide range of problems across all aspects of an e-commerce platform at Meesho. With around 120 million active users, Meesho is one of India’s fastest-growing e-commerce platforms. Open-sourcing their ML platform with proven capabilities in handling high volumes of requests is a significant step towards empowering the global AI community.
