A quiet race is on to give India its own AI voice. From call-centre automation to defence and legal systems, Abhishek Upperwal of Soket Labs and journalist Swathi Radhakrishnan tell us why AI trained only on Western, English-heavy data cannot meet India’s needs. Translation isn’t enough; models must “think” in Hindi, Tamil or Marathi to capture nuance and reduce bias. The government’s IndiaAI mission, with nearly ₹10,000 crore in funding, is catalysing startups to build these Indic models. But their biggest bottleneck is data. Only a sliver of the world’s open datasets are in Indian languages, and even public archives like Doordarshan take time to unlock. Startups are scrambling crowdsourcing voices, licensing publishing-house content, generating synthetic text and negotiating with ministries to reach the 15–20 trillion high-quality tokens needed for a world-class foundation model. In this episode Host Anirban Chowdhury, ET’s Swathi Moorthy and Soket AI’s, founder, Abhishek Upperwal try to answer the following questions:
- What makes sovereign, Indic-first AI critical for India’s economy and security?
- How are innovators overcoming the huge shortage of quality language data?
- Can low-cost, DeepSeek-style methods help India build frugal yet powerful models?
- Where will the commercial payoffs voice AI, regional apps, enterprise tools arrive first?
Tune in.
You can follow Anirban Chowdhury on his social media: Twitter and Linkedin
You can follow Swathi Moorthy on her social media: Twitter and LinkedIn and also read Newspaper Article
Listen to Corner Office Conversation our new show:: Corner Office Conversation with Pawan Goenka, Chairman, IN-SPACe, Corner Office Conversation with The New Leaders of Indian Pharma and much more.
Check out other interesting episodes from the host like: Why Is India Still Buying Russian Oil?, How AI is Rewriting Cinema Part 2, Trump vs Harvard: India Impact, Of Dragons and Elephants: Modi–Xi in Focus and much more.
Catch the latest episode of ‘The Morning Brief’ on ET Play, The Economic Times Online, Spotify, Apple Podcasts, JioSaavn, Amazon Music and Youtube.
See omnystudio.com/listener for privacy information.
