4 billion payments. 3 trillion data points. One model trained on all of it.
Meet @Razorpay Vulcan - India’s first transformer-based AI foundation model for payments.
Built, trained and hosted in India, in partnership with @nvidia and @awscloud.
Until now, every payments
20 weeks in. The thesis is holding: dashboards are dying.
The interface for software is moving from click, search, filter to simply ask.
On Razorpay’s AI-first dashboard (still in Beta):
→ 7.8x weekly queries
→ 5.3x new merchants/week
Get early access:
Lot of debates on this shift.
Sharing our learnings from building @Razorpay's AI-first dashboard.
<5% of B2B users open a dashboard to explore - most come with a question, but spend time figuring out how to get the answer.
AI-first flips this: just ask for what you need.
Been playing with 2 DGX Sparks for the last few weeks, running DeepSeek V4 Flash 0731 official FP8.
Now running a near Opus-level model locally at:
→ 75-89 tok/s single stream (TP=2 mode)
→ 1M+ context (verified retrieval)
Significant upfront cost today, but this feels like a
Witness the greatest redemption story ever told.🥊
The story of a fallen champion.
Written off, replaced, forgotten in the age of one-tap.This monsoon, the old trusted guard returns - sharper, wiser and battle-ready
NETBANKING - BACK IN THE RING 🥊
Coming soon to your