AI products built for India must balance model flexibility, affordability and local user needs, as technology leaders adapt systems for diverse devices, languages and connectivity.
The challenge of taking AI products to millions of Indian users is making them work across budget smartphones, multiple languages, and patchy networks while keeping costs in check. At Inc42’s inaugural ‘The CTO Summit 2026’ in Bengaluru, executives discussed how they are tackling these barriers to wider adoption.
Speaking during the session, “Building AI Systems For Indian Scale”, executives from ShareChat and Moj, Meesho, Rapido, and Shadowfax outlined their approaches, from voice-led shopping assistants to tools that help delivery workers locate customers.
Rapid changes in AI models add another challenge: companies need infrastructure that allows them to switch providers without rebuilding their applications.
Nitin Jain, CTO at ShareChat and Moj, said companies should avoid becoming tied to a single model or approach. “If we get married to one particular kind of model or approach, that’s not going to work,” he said.
ShareChat focuses on infrastructure that lets teams replace models, reuse contextual information, and assess performance when switching providers, he added.
Jain also outlined how India-specific constraints have shaped the company’s technology. Its users often rely on mid-range phones and limited bandwidth, while the business needs to support multiple languages and deliver a competitive experience despite lower monetisation levels. These requirements have driven investments in data processing and recommendation systems designed to serve a large user base while keeping costs in check.
Making Product Discovery More Accessible
At Meesho, adapting to users' needs has included developing Vaani, a voice-led shopping assistant that helps shoppers express what they want to find.
Anand Jain, head of engineering at Meesho, said the feature was designed to help users, particularly those with limited digital experience, navigate product discovery.
Meesho reported 22% higher conversions among Vaani users than non-users in some Tier III and IV cohorts. The feature was activated by 1.5 Mn users in its first month, he said.
Notably, the focus on understanding shoppers’ intent also featured in Flipkart’s session at the summit earlier in the day, with CPTO Balaji Thiagarajan outlining how the company is combining search, conversational interactions, voice, and visual inputs to personalise product discovery.
Meesho is also exploring open-weight models and testing on-device AI to manage the cost of serving a large user base. Jain said the company is testing how processing AI requests on phones affects memory, battery life, and the user experience, particularly on lower-end devices.
Keeping AI Costs In Check
Rishikesh SR, cofounder of Rapido, said the company has built software layers that allow teams to work with different models and established rules for their use, while giving employees access to advanced AI models.
He put the combined model acquisition and people costs at less than four paise per ride, stressing that experimentation needs accountability as usage grows.
At Shadowfax, AI is being used to address a longstanding logistics problem: matching incomplete or inaccurately written addresses to precise locations.
Vaibhav Khandelwal, cofounder and CTO of Shadowfax, said the company uses historical delivery data and AI-based address matching to improve location identification. Incorrect pincodes and unclear addresses can delay deliveries and increase logistics costs, he said, putting the impact at around 1% of the company’s margins.