From typos to Hindi phrases, Zepto says its search now understands everyday language inputs
Zepto is retooling how people find products in its app, saying its search now understands the way Indians actually speak and type—typos, Hinglish, and regional phrases included. Co-founder Kaivalya Vohra shared that the company’s new system aims to catch intent even when users don’t use formal product names, a common gap that has long frustrated shoppers across ecommerce apps.
Designing for how India really searches
In a LinkedIn post, Vohra noted that many users won’t type “laundry basket,” but will instead write “kapde rakhne ki bag”—a perfectly understandable request in everyday Indian parlance. Historically, such queries often returned no results or irrelevant items because catalogs and category trees were built around English product names and standardized keywords.
That disconnect prompted Zepto’s team to rethink search from the ground up. According to Vohra, the company has built a system that can parse regional language inputs, handle spelling variations, and infer user intent. Under the hood, he said, it’s powered by custom large language models tailored to how users in India think and type, rather than relying on literal translations.
From keywords to intent
The shift echoes a broader industry realization: people don’t think in keywords; they think in goals. Searching for “kapde rakhne ki bag” signals a clear need, even if it doesn’t match a catalog’s English-centric nomenclature. When search doesn’t bridge that gap, discovery suffers.
One commenter summed it up bluntly: “Site search in ecommerce is almost entirely a synonym and intent problem. Catalog taxonomy gets built around English SKU names and standard category trees, which is exactly not how Bharat searches.” The result, they added, is that high-intent queries often show zero results—not due to stock issues, but because catalogs are structured for a different type of user.
Another user echoed the point: “Intent-based search is the need of the hour. Users don’t think in keywords, they think in intent.” In large inventories, over-reliance on keyword mapping can bury relevant products, creating missed discovery opportunities and, ultimately, lost conversions.
How Zepto says its system works
- Understands everyday language: Recognizes colloquial and regional phrasing such as Hinglish or Hindi queries that don’t mirror official product names.
- Handles typos and variations: Accounts for misspellings and non-standard spellings that are common on mobile keyboards.
- Focuses on intent: Interprets what a user is trying to accomplish, not just the exact words they type.
- Built on custom LLMs: Uses large language models adapted to Indian user behavior, rather than direct translation approaches.
Vohra framed this as a philosophy shift: design for Indian users as they are, not as catalogs assume them to be. It’s a move away from rigid keyword matching and toward a semantic layer that interprets meaning.
Why it matters for quick commerce
In rapid-delivery apps, speed isn’t just about logistics—it begins with how fast and accurately users can locate what they want. If someone can type a phrase in their natural language and get precisely what they need, that shortens the path to purchase and reduces friction. For Zepto, which competes in a crowded quick commerce market, tightening that loop could translate into better conversion and retention.
Part of a broader product push
Founded in 2021 by Vohra and Aadit Palicha, Zepto operates in the quick commerce segment, delivering groceries and daily essentials at speed. This year, the company has rolled out new in-app features such as a pay-later option, while continuing to invest in supply chain automation and operations. It has also been preparing for a public market debut as competition intensifies across the category.
The takeaway
Zepto’s search overhaul targets a long-standing blind spot in ecommerce: English-first catalogs that don’t align with how a diverse set of users actually describe what they need. By leaning into regional language inputs, tolerating typos, and prioritizing intent, the company says it’s building a search experience that feels native to Indian shoppers. If it works as described, it could turn everyday language into a first-class interface for discovery—and make quick commerce feel a bit more, well, quick.
