Building AI4Bharat: Why India Needs Its Own AI Stack, With IIT Prof. Mitesh | Decoding AI Episode 8
Автор: SeedToScale
Загружено: 2025-07-10
Просмотров: 1735
India’s AI future won’t be built by copying Western models. It needs infrastructure rooted in Indian languages, accents, and real-world use cases. At the heart of this mission is AI4Bharat, an open-source initiative building cutting-edge AI models for all 22 official Indian languages.
In this episode of Decoding AI, Prayank Swaroop, Partner at Accel, sits down with Professor Mitesh Khapra, one of the key architects of India’s Indic and open-source AI movement and co-lead of the AI4Bharat initiative at IIT Madras. A former researcher at IBM and now a faculty member at the Wadhwani School of Data Science and AI, IIT Madras, Mitesh has been instrumental in laying the groundwork for India’s AI stack.
The conversation unpacks what it means to build open-source AI infrastructure for Bharat. Prayank and Mitesh discuss India’s foundational gaps in data, why off-the-shelf global APIs aren’t ideal in the long run, and how open-source models trained on Indian languages can unlock a new wave of inclusive AI innovation.
Chapters:
01:00 Prof. Khapra’s move from industry to academia
04:02 Early triggers behind AI4Bharat
05:28 AI4Bharat's discovery-first approach to problem-solving
08:23 Early use case: Reading ECG scans
11:46 Data collection roadblock
13:25 Pivot to language technology
16:24 Why Indian startups default to foreign APIs
19:06 Democratization of AI, and the 3 components of AI
21:04 Building self-improving data loops
22:34 Synthetic data for voice training
24:07 What matters when building AI for India
27:44 Mainstream versus India-specific focus
28:19 Code-switching and colloquialization
30:58 How AI4Bharat collaborates with startups
32:36 Developer evangelism
33:50 How to select the right AI4Bharat model for your use case
37:24 How AI4Bharat's models compare to global standards
38:12 Voice actors' role in speech synthesis quality
40:30 The 5-year vision for India’s AI ecosystem
42:19 Conclusion
Explore AI4Bharat’s open-source models: https://models.ai4bharat.org/
More episodes of Decoding AI: • Decoding AI
Follow SeedToScale for more founder-first content and conversations:
SeedToScale on LinkedIn: / seedtoscale
SeedToScale on X: https://x.com/Seed2Scale
Follow Mitesh Khapra: / mitesh-khapra-3bb3032
Follow Prayank Swaroop: / prayank
About Decoding AI
Decoding AI delves into the real questions surrounding the building, deployment, and scaling of AI through conversations with founders and operators at the forefront of AI.
Subscribe to @seedtoscale for more insights, interviews, and founder-first content that help you build with clarity and conviction.
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке: