Hypervalue ecommerce platform has open-sourced the key components of its internal machine learning platform, BharatMLStack, as it aims to make AI infrastructure more accessible to developers. The release includes its online feature store, control plane, orchestration UI, and SDKs—all now available on GitHub.
Last year, Debdoot Mukherjee, Head of AI and Demand Engineering at Meesho, had announced the Prosus-backed company’s plans to open-source its ML platform, including the feature store and model orchestrator.
The IPO-bound social ecommerce company said its BharatMLStack—developed over the last two years—powered over 66.9 trillion feature retrievals (data signals fetched to make real-time predictions) and 3.12 trillion real-time inferences (real-time predictions) at peak during FY25.
Meanwhile, its online feature store serves precomputed, real-time features to machine learning models and has essential use cases like fraud detection and personalised recommendations.
Meesho said BharatMLStack can drastically reduce the time and complexity involved in deploying ML solutions. “We put BharatMLStack to the test during high-traffic events like our Mega Blockbuster Sale in March 2025, where it delivered at scale, demonstrating its ability to perform under peak load conditions,” Sanjeev Kumar, Founder and CTO of Meesho, said in a statement.
He added, “This helped Meesho drive higher user engagement, better conversions, and increased order volumes during the sale. By open-sourcing it, we’re sharing a high-scale, AI stack with the broader tech community—purpose-built for real-time use cases and tailored for Indian businesses.”
The company said future components of the stack will also be open-sourced, with opportunities for the developer community to contribute and iterate together.
