In a move aimed at democratizing access to advanced AI infrastructure, Meesho features the open source key components of Bharatmlstack, an internal machine learning (ML) platform on GitHub.
This release, which includes a production-grade feature store, control plane, orchestration UI, and SDK, places IPO bound companies among the first major Indian ecommerce players to contribute to the growing AI ecosystem at this scale.
Developed over the past two to three years, Bharatmlstack was dedicated to supporting a wide range of real-time machine learning applications across systems heading towards Meesho users and sellers. In 2025, the platform processed an average of 1.91 petabytes of data per day, performed a search for 66.9 trillion features, and enabled 3.12 trillion real-time inference during peak loads.
The initiative stems from the company's decision to unify ML's efforts to a central and efficient system that can handle real-time use cases without reducing operational costs.
Meesho's AI leadership highlighted that the platform was rigorously tested during Megablockbuster sales in March 2025, successfully handling traffic surges while improving engagement, conversions and order volumes.
“Good technology needs to expand impact not only infrastructure, but also infrastructure,” Sanjeev Kumar, founder and CTO of Meesho, said in a statement.
“We tested and delivered on a massive scale at a busy event like the Mega Block Buster Sale in March 2025. We demonstrated our ability to run under peak load conditions. Cases and customizations for Indian companies.”
Beyond raw performance, Bharatmlstack is designed to accommodate unique Indian internet behaviors such as transliterated search queries, low resource devices, and fuzzy products discovery. Its modular components are built to support use cases such as fraud detection, personalized search, ranking systems, and automated product tagging.
Meesho's chief data scientist, Debdoot Mukherjee, explained that the company initially relied on a combination of open source and its own ML tools, but found it to be too expensive or inflexible.
“With our own platform, all we need is a compact vehicle, like driving a sports car when we can't even look under the hood,” he said. MoneyControl. “Open source provides transparency and control.”
Meesho's open source release starts with an online feature store. This starts with a tool that provides pre-computed functionality in real-time to the ML model. Such stores are important to ensure that systems such as product recommendations and fraud detection work with low latency and high consistency, avoiding data silos between teams.
Empowering Indian AI builders
With this release, Meesho aims to equip the Indian AI and data science community, including early stage startups and independent ML engineers, with tools that are not only scalable but also financially sustainable.
The company is not trying to monetize its release at this stage. Instead, the objective is to promote collaboration, invite feedback, and crowdsource improvements.
Mukherjee noted that early stage startups without dedicated platform teams could benefit from Bharatmlstack's plug-and-play architecture, but more mature companies that hit scale can build robust and cost-effective infrastructure without reinventing the wheel.
Additional components such as models, model registration systems and workflow authoring tools that provide infrastructure are expected to be open sourced in stages over the coming months. Meesho says even “12 companies” think the initiative will be successful if they significantly integrate the stack into the core ML workflow.
