In a bold move to democratize AI in India, Meesho has open-sourced key components of its internal ML platform, BharatMLStack, on GitHub. The IPO-bound e-commerce giant now become one of the first major Indian startups to contribute open-source AI infrastructure at this scale, specifically designed for real-time, high-traffic use cases common in India’s digital economy.
Built over the past 2–3 years, BharatMLStack powers several of Meesho’s core features, ranging from personalized recommendations to fraud detection. In FY2025 alone, the system processed an average of 1.91 petabytes of data daily, supported 66.9 trillion feature retrievals, and enabled 3.12 trillion real-time inferences during peak load events.
The platform was rigorously tested during Meesho’s Mega Blockbuster Sale in March 2025, where it scaled seamlessly to handle massive user traffic, improve engagement, increase conversions, and drive higher order volumes.
Meesho’s CTO, Sanjeev Kumar, emphasizes the importance of scalable AI infrastructure in enhancing the impact of great technology beyond infrastructure.
What Is Meesho’s BharatMLStack?
BharatMLStack is a modular, production-grade machine-learning platform that includes:
- A real-time online feature store
- Orchestration UI and SDKs
- The control plane for workflow management
Its design accommodates Indian internet conditions—handling transliterated search, low-resource devices, and fuzzy product discovery.
Initially built using a mix of proprietary and open-source tools, Meesho pivoted to a custom solution after finding commercial platforms too costly or inflexible.
Meesho’s Chief Data Scientist, Debdoot Mukherjee, compared using proprietary tools to driving a sports car to a compact car, stating it’s difficult to open the hood.
Why It Matters for Indian Startups
The release is aimed at helping early-stage startups, ML engineers, and mid-sized companies build scalable ML workflows without massive infrastructure budgets.
Meesho’s decision not to monetize the platform reflects a broader goal: collaboration, transparency, and crowd-sourced innovation. The plug-and-play architecture makes it easier for startups with limited platform teams to adopt and integrate the tools.
By addressing the unique challenges of Indian digital commerce and user behavior, BharatMLStack lowers the barrier to entry for companies seeking real-time ML capabilities without needing to reinvent the wheel.
What’s Coming Next?
Meesho plans to open-source additional components in phases:
- Model serving infrastructure
- Model registration systems
- Workflow authoring tools
If even “a dozen companies” adopt BharatMLStack in production, Meesho will consider the initiative a success.
In a market where cutting-edge ML tools are often out of reach for smaller firms, Meesho’s open-source move could be a game-changer, setting the stage for a more inclusive, innovation-driven AI ecosystem in India.
