ABOUT THIS EPISODE
Everyone is watching the frontier-model race. Bhavik Vasa thinks India's actual opportunity is somewhere far less glamorous: small, domain-specific models sitting on top of proprietary Indian data, quietly turning a ten-day underwriting process into a ten-minute one.
Bhavik Vasa is a senior leader at GetVantage, an India-based embedded-finance company lending to small businesses — in the episode he speaks for its credit, data-science and engineering teams. He argues that standard LLMs trained on the open internet are becoming a commodity, that the unit economics of running them have not played out yet, and that the real economic value is shifting to applied AI at the enterprise level.
We talk about why India shouldn't try to build another trillion-parameter model, what UPI, GST and the India Stack actually unlocked for credit, why a digital footprint is not the same thing as a credit decision, why the MSME lending explosion has taken so long to arrive, how GeM Sahay turns an approved government invoice into instant credit for the small businesses selling to the government on GeM — ten lakh of them, by his count, and why he thinks financing is about to become as invisible as payments already are.
Chapters:
0:00 Cold open
0:46 Has the AI revolution reached B2B?
2:32 We've seen this hype cycle before
3:53 The value shifts to applied AI
6:32 Why India should build SLMs, not another frontier model
9:03 Instant underwriting on proprietary Indian data
10:33 What LLMs changed that machine learning couldn't
12:45 UPI, GST and the India Stack
13:54 A digital footprint is not a credit decision
17:51 If the rails are ready, where is the lending explosion?
21:53 Does regulation slow this down?
25:20 Squaring SLMs with OpenAI and Anthropic's growth
26:38 "You're only looking at revenue, not the cost of operations"
29:02 Is enterprise data actually ready?
30:48 GeM Sahay: instant credit for ten lakh MSMEs
34:30 The future of financing is invisible
35:40 The AI tools Bhavik actually uses
38:18 Embedding credit inside Tally and QuickBooks
40:24 Closing thoughts
