Meesho Open-Sources BharatMLStack to Power India’s Next Generation of AI Startups

by Shan 2025-07-03 10:59:52

India’s AI startup scene just got a big boost. Meesho, one of the country’s most active digital commerce platforms, has just released something it’s been quietly building for years: BharatMLStack—a set of internal tools that handle its massive real-time machine learning operations. And here’s the best part: it’s now open source.

That means Indian developers, small startups, and even college projects can now tap into the same tools Meesho uses to process 1.91 petabytes of data and serve over 3 trillion predictions—for free.

What Exactly Is BharatMLStack?

Think of BharatMLStack as the toolkit behind Meesho’s AI engine—used to run personalisation, fraud detection, pricing models, and more. These tools were built in-house to handle real-time machine learning at scale. Now, they’re on GitHub for anyone to use.

Here’s what’s inside:

Component	What It Does
Feature Store	Stores precomputed data points (called "features") to speed up ML models
Orchestration Interface	Helps run and monitor ML workflows efficiently
Control Plane	Manages model versions, deployments, and monitoring
SDKs	Developer tools to plug ML features into apps

These aren’t beta-stage experiments. These tools supported Meesho during high-traffic events like the Mega Blockbuster Sale in March 2025, handling millions of transactions with near-zero latency.

Why Does This Matter?

Let’s be honest—building AI infrastructure is hard. It takes time, money, and lots of trial and error. Most startups don’t have that kind of runway.

That’s why Meesho’s decision to open up its production-grade stack is so important. Here’s what it unlocks:

1. Less Time Spent on Infrastructure

Startups can now skip building the plumbing and jump straight into developing real use cases. Whether it’s fraud detection, churn prediction, or pricing models, you can focus on your models—not managing data pipelines.

2. Built for Real-Time

Most off-the-shelf open-source tools weren’t made with real-time systems in mind. BharatMLStack was. It's designed to handle billions of feature lookups and real-time predictions. That’s a huge deal for anyone building:

Instant credit scoring
Live recommendation engines
Flash-sale optimisation tools
Real-time logistics or delivery ETAs

3. Made for India’s Scale

Meesho didn’t build BharatMLStack for theory—it built it to serve millions of users in Tier 2 and Tier 3 cities, often on unstable networks and low-powered phones. That makes it more aligned with India-first startups than global solutions that need heavy localisation.

Real Example: BharatMLStack During the 2025 Mega Sale

Let’s take a quick look at how Meesho used BharatMLStack during its March 2025 Mega Blockbuster Sale:

Use Case	Impact
Personalised feeds	Served in milliseconds based on real-time user behaviour
Real-time fraud detection	Blocked suspicious activity as it happened
Dynamic pricing adjustments	Updated pricing as inventory moved
Logistics & ETA optimisation	Improved delivery estimates using live data

That’s the same infrastructure that’s now public.

Who Can Use This Stack?

If you’re building anything that uses data to make decisions in real-time, this release is for you.

Who	Why This Helps
Startup founders	Fast-track MVPs without building backend infra
Data scientists & ML engineers	Train and serve models quickly using production-ready tools
Product managers	Add AI features without reworking the entire system
Students & open-source devs	Learn by working on real infra, not toy datasets

What’s Live on GitHub?

Meesho has published:

The Feature Store – the same tool it uses to store and serve features across teams
Control Plane APIs – to manage model deployment and lifecycle
Workflow orchestration layer
Software Development Kits (SDKs)
Documentation + examples

More modules will be released in phases. Meesho is also inviting contributions—so if you’re someone who loves building tools for scale, this could be your chance to make a mark.

Why Now? A Strategic Move After Meesho’s Reverse Flip

This launch also comes right after Meesho completed its reverse flip back to India—moving its holding structure from the US to India, with the NCLT greenlighting the final step.

In plain words: Meesho is doubling down on being a tech-first company built in India, for India. Making BharatMLStack public shows it wants to bring the community along for the ride.

What Meesho Says

“These tools helped us scale machine learning across Meesho. By sharing them, we want to make it easier for others to build fast, smart, AI-driven products. This is our way of giving back to the community.”
— Spokesperson, Meesho

How to Get Started

If you’re curious about trying out BharatMLStack, here’s a quick starter path:

Go to the GitHub repo and check out the docs
Set up the feature store locally using your own dataset
Run the sample workflows for a pricing or fraud detection model
Plug into your existing ML pipeline or app backend
Give feedback or contribute new modules—Meesho is listening

Key Takeaways

BharatMLStack is now open-source: Meesho’s internal AI stack is public for developers and startups
It’s built for scale: Used in production for 3 trillion+ predictions and massive sales events
It’s made for real-time use: Perfect for fraud detection, personalisation, and pricing
It’s built in India, for India: Works well with Indian user behaviour and data scales
More modules are coming: Feature store and orchestration tools are just the start

Final Thought

Meesho isn’t just releasing code—it’s sharing battle-tested tools that can help every small team think and build like a big tech company. BharatMLStack lowers the entry barrier for AI infrastructure, and that’s a massive win for India’s developer and startup ecosystem.

If you’ve ever wanted to build serious AI without breaking your brain—or your budget—this might be your moment.