![](https://worldnewsradar.id/wp-content/uploads/2025/08/BANNER-RENT-300X47-COLUMN-1.png"){kind=small}
In a landmark move to democratize access to scalable AI infrastructure, Indian e-commerce platform Meesho has open-sourced critical components of its proprietary machine learning (ML) platform, BharatMLStack, on GitHub. This strategic decision places Meesho among the first major e-commerce players in India to contribute this extensively to the nation’s AI development landscape. With this initiative, Meesho aims to enable startups, independent developers, and established companies to build, deploy, and manage real-time AI models efficiently and affordably.
I. A Technological Leap for the Indian AI Community
1. Unveiling BharatMLStack
After more than two years of development, Meesho has released BharatMLStack—its high-performance ML platform powering real-time applications across user and seller interfaces. The open-sourced components include a production-ready feature store, orchestration UI, software development kits (SDKs), and a control plane, making it a comprehensive infrastructure solution for advanced AI use cases.
2. Built for Scale and Local Relevance
Designed to meet the demands of Indian users and platforms, BharatMLStack is optimized for real-time inference and massive-scale data handling. In FY2025 alone, it processed approximately 1.91 petabytes of data daily, facilitated 66.9 trillion feature retrievals, and supported 3.12 trillion real-time predictions during peak traffic periods. It has been specifically engineered to accommodate local internet behaviors, such as transliterated searches, low-spec devices, and fuzzy product searches.
II. Designed for Performance, Efficiency, and Affordability
1. Centralizing ML Operations
BharatMLStack is the result of Meesho’s initiative to streamline its machine learning operations into a unified, efficient, and scalable platform. By consolidating tools and processes, the company eliminated the high operational costs associated with fragmented ML systems and ensured real-time capabilities for critical applications like personalized recommendations, fraud detection, and product ranking.
2. Successfully Stress-Tested at Scale
The robustness of the platform was validated during Meesho’s Mega Blockbuster Sale in March 2025. BharatMLStack handled record-breaking traffic without performance hiccups, while simultaneously enhancing user engagement, improving conversion rates, and boosting overall order volumes. According to Meesho’s Founder and CTO, Sanjeev Kumar, this real-world validation confirmed the platform’s scalability and effectiveness.
“Great technology should scale impact, not just infrastructure,” Kumar stated.
III. Transparency Over Proprietary Complexity
1. The Shift Away from Costly, Closed Systems
Initially, Meesho experimented with a blend of open-source and proprietary ML tools but found that closed systems lacked flexibility and transparency. According to Debdoot Mukherjee, Meesho’s Chief Data Scientist, proprietary tools were overpowered for many use cases and didn’t allow full visibility into their inner workings—limitations that open-source systems easily overcome.
“With proprietary platforms, it’s like driving a sports car when all you need is a compact vehicle and you can’t even look under the hood,” Mukherjee explained.
2. Leveraging Open Source for Customization and Control
By opting for open-source infrastructure, Meesho ensures greater control, adaptability, and community-driven innovation. BharatMLStack’s modular and plug-and-play design allows developers to pick and integrate components based on their specific needs—be it for recommendation engines, fraud prevention, or tagging automation—without getting locked into rigid or costly software ecosystems.
IV. First Release: Online Feature Store
1. Real-Time Feature Availability
The first component released under the BharatMLStack initiative is Meesho’s Online Feature Store. This system allows ML models to retrieve precomputed features in real time, ensuring that applications maintain low latency and high accuracy. Whether it’s for fraud detection or personalized product recommendations, access to consistent and fast data is crucial for real-time decisions.
2. Breaking Down Data Silos
The feature store eliminates the issue of data silos by providing a centralized, team-accessible environment for feature storage and retrieval. This improves consistency across model training and production environments while enhancing collaboration among engineering and data science teams.
V. Empowering India’s Next Generation of AI Innovators
1. Supporting Startups and Independent Engineers
With no immediate plans to monetize the open-source platform, Meesho’s primary goal is to support the broader AI and data science community in India. BharatMLStack offers a ready-to-use solution that can help startups and solo developers build and deploy machine learning systems without investing in costly platform engineering.
2. Encouraging Feedback and Community Growth
The company hopes to crowdsource improvements by inviting feedback and collaboration from users. Mukherjee emphasized that even if just a dozen organizations implement BharatMLStack meaningfully into their ML workflows, the initiative would be considered a success.
3. Roadmap for Further Releases
In the coming months, Meesho plans to expand the open-source offering by releasing additional tools, including model serving infrastructure, model registration systems, and workflow authoring tools. These additions will provide end-to-end support for ML lifecycle management—from data ingestion and model training to deployment and monitoring.
Conclusion
Meesho’s decision to open-source BharatMLStack represents a meaningful step forward for India’s AI and tech ecosystem. By releasing production-grade ML infrastructure tailored to local needs and real-time applications, the company is fostering an environment of transparency, collaboration, and accessible innovation. With no commercial strings attached, Meesho is paving the way for a new wave of AI builders in India—making high-scale machine learning development not only possible but also practical and affordable.
