Key Points
- Meesho is open-sourcing its ML platform at the Nvidia Summit.
- The initiative includes a feature store and model orchestrator.
- Meesho’s system can handle millions of requests every second.
- Generative AI makes it easier for sellers to list products and translate content.
- Meesho is working with IISc to advance research in generative AI.
Looma News
At the Nvidia Summit in Mumbai, eCommerce giant Meesho announced that it is open-sourcing its machine learning platform, including its feature store and model orchestrator. Debdoot Mukherjee, the head of AI at Meesho, highlighted that this initiative will enable developers and data scientists to use AI more effectively.
During his talk, “Cost Effective High-Performance Model Serving for the Masses,” Debdoot shared insights about Meesho’s large-scale machine learning models. He talked about the Feature Store, which is built for high-performance machine learning operations. This store helps manage and standardize features at scale and can handle millions of requests every second, with 99 percent of requests processed in under 10 milliseconds. Debdoot also confirmed that the Feature Store will be open-sourced.
In a LinkedIn post, he mentioned their progress in scaling transformer-based ranking models in production, highlighting improvements made with TensorRT to meet both speed and cost needs. Recently, at the Build With Meta Summit, it was announced that Meesho plans to integrate Meta’s open-source Llama models into its technology stack.
Meesho has been actively using generative AI and machine learning to enhance its platform. Its large-scale decision engine improves code maintenance, reduces delays, and offers better error handling with multiple backup options. This engine also lets developers monitor system health through a dashboard for easy troubleshooting.
For users, Meesho uses generative AI to make the listing process smoother for sellers. It can translate product descriptions, addresses, and voice searches into various Indian languages besides Hindi. Additionally, Meesho is using Microsoft’s Azure OpenAI to improve call center operations, reportedly achieving a 20% boost in customer satisfaction scores.
Last year, Meesho partnered with IISc to further research in generative AI and develop multimodal representation learning. Debdoot stated, “At Meesho, AI and ML are essential to solve a wide range of issues across every part of our e-commerce platform.” With around 120 million active users, Meesho’s decision to open-source its robust machine learning platform could greatly benefit the ecosystem, helping developers create similar ML systems and tackle existing tech challenges.