logo
#

Latest news with #BharatMLStack

Meesho open-sources BharatMLStack to boost AI innovation in startups
Meesho open-sources BharatMLStack to boost AI innovation in startups

Time of India

timea day ago

  • Business
  • Time of India

Meesho open-sources BharatMLStack to boost AI innovation in startups

In a move aimed at unlocking access to AI infrastructure, Meesho has open-sourced key components of its internal machine learning platform, 'BharatMLStack,' on GitHub — including its feature store, control plane, orchestration UI, and SDKs. With this initiative, Meesho is among the first scaled e-commerce players in India to take a step forward toward AI development. Developed over the past two years, BharatMLStack, Meesho's Machine Learning Platform, processes large volumes of data every day. In the financial year 2024- 2025, our machine learning systems processed ~1.91 PB data daily on average in Fiscal 2025, resulting in 66.90 trillion features retrievals (data signals fetched to make real time predictions) and 3.12 trillion inferences ( real-time predictions made by ML models) at peak. 'We believe great technology should scale impact, not just infrastructure. 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. This helped Meesho drive higher user engagement, better conversions, an d 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.' , said Sanjeev Kumar, Founder & CTO, Meesho. An Online Feature Store helps AI systems by serving precomputed, up-to-date features instantly for real-time model predictions, enabling use cases like fraud detection and personalized recommendations to run faster and more accurately. By sharing this with the broader tech ecosystem, Meesho aims to empower startups, ML engineers, and data scientists across India with access to production-ready, cost-efficient AI tooling tailored for India's fast-evolving digital economy. Teams can collaborate more effectively, avoid data silos, and quickly build or enhance AI models with consistent, high-quality features, all while accelerating deployment and driving innovation. While it doesn't replace model development, it significantly simplifies training, deployment and infrastructure management, allowing data scientists and engineers to focus on what matters most: building better models and accelerating innovation.

AI for the masses: Inside Meesho's bold leap with BharatMLStack
AI for the masses: Inside Meesho's bold leap with BharatMLStack

Time of India

time3 days ago

  • Business
  • Time of India

AI for the masses: Inside Meesho's bold leap with BharatMLStack

Artificial Intelligence is no longer a back-end tool—it sits at the very heart of digital commerce, shaping how technology communicates, recommends, and influences consumer decisions. Underlining this, a report titled EY Consumer Index Survey sheds light on India's readiness for this shift: 82% of Indian consumers believe AI will significantly improve their shopping experience, while 30% say they understand AI, compared to just 17% globally. Surprisingly, despite being an emerging tech market, Indian consumers are outpacing their global counterparts in AI awareness and trust—especially in areas like personalized recommendations, where 48% of Indians trust AI to tailor deals and offers, double the global average. To explore how AI is being scaled, democratized, and personalized for 'Bharat,' ETCIO interacted with Debdoot Mukherjee, Head of AI & Demand Engineering at Meesho , one of India's fastest-growing e-commerce platforms . Edited Excerpts: Meesho recently open-sourced the BharatMLStack on GitHub, becoming one of the first Indian e-commerce platforms to do so. What strategic vision underpins this move, and how do you see it shaping India's AI ecosystem, especially in Tier 2+ markets? The decision to build and eventually open-source BharatMLStack came from our need to develop a robust internal ML platform to support hundreds of diverse use cases at scale. We needed something cost-efficient, reliable, scalable, and extensible—and over time, our in-house stack evolved to serve 66.90 trillion of feature retrievals and real-time predictions. That's when we decided to open-source BharatMLStack—to help startups and larger companies leapfrog their AI journeys. It's especially useful for those building for Bharat, where scale and cost efficiency are critical. On that note, how is Meesho leveraging AI to power smarter, more intuitive shopping experiences for mobile-first, value-conscious users in Bharat? Can you share some use cases that show direct impact? Absolutely. For users from Tier 2+ towns, there are barriers to online shopping—language, ambiguity in search, and lack of digital familiarity. AI helps break these barriers. One major area is personalization—we use deep learning models across surfaces like homepage feeds and search to serve relevant products based on user behavior and intent. On the seller side, our AI helps ease cataloging through image recognition and enrichment, automates pricing insights, and improves product visibility—especially critical for first-time or non-tech-savvy sellers with razor-thin margins. Are there any unique challenges or breakthroughs your team has encountered while scaling these AI systems? Yes, several. One major challenge is interpreting long-tail, ambiguous search queries, often typed in Indian languages using English scripts or even voice. We've used LLMs to decode intent and improve search relevance. Another is balancing relevance with fairness—ensuring that newer sellers also get visibility despite algorithmic biases like the 'rich get richer' effect. Our ranking systems are designed to support platform fairness, long-term seller success, and ultimately, greater consumer choice and price competitiveness. With AI becoming more influential in consumer apps, how do you approach responsible AI, bias mitigation, and compliance, especially in light of India's evolving data privacy regulations? Data privacy is something we take very seriously—we're fully aligned with current and upcoming regulations. But beyond compliance, responsible AI is embedded into our philosophy. We constantly monitor for algorithmic bias, especially in recommendation systems, and work to counteract unintended patterns that could marginalize certain sellers. This isn't just good ethics—it's good business. A healthy, competitive seller base offers better prices and more variety, benefiting the platform and our users. Looking ahead, what are some areas of AI innovation you and your team are most excited about—LLMs, multimodal models, or deeper integrations in logistics, seller support, and beyond? There's so much happening, it's hard to choose! But yes, GenAI and LLMs are becoming central. We've implemented GenAI-powered customer support chatbots and are expanding that across user categories. In logistics, we've built an AI-first platform called Valmo, where everything from delivery planning to fraud detection is AI-powered. For instance, we created a Geo-India LLM to decode vague, rural delivery addresses—something no other off-the-shelf tool could do. It's a great example of multimodal intelligence tailored for India.

DOWNLOAD THE APP

Get Started Now: Download the App

Ready to dive into a world of global content with local flavor? Download Daily8 app today from your preferred app store and start exploring.
app-storeplay-store