UPI is the poster baby of India’s tech innovation and a benchmark for issues to return. Its easy design and the sheer functionality to scale it to a billion customers make it a masterpiece—one which the world has been making an attempt to copy in useless.
Behind UPI’s seamless simplicity lies a narrative of hard-won technological independence, system-level considering, and relentless problem-solving. On the centre of this story is Vishal Anand Kanvaty, CTO at NPCI, and one of many key architects of UPI.
Kanvaty opened up with AIM concerning the evolution of NPCI’s know-how journey and the foundational selections behind UPI’s structure. “We had three phases at NPCI,” Kanvaty mentioned, reflecting on the organisation’s tech evolution.
Within the first section, every little thing from product design to operations was vendor-managed. They used to run the product and the platform, and NPCI used to simply handle the distributors.
The second section introduced marginal management, with NPCI shopping for merchandise from distributors like Oracle and Euronet however nonetheless relying on them for enhancements. This mannequin created bottlenecks. “NPCI at the moment was comparatively small, and Indian prospects weren’t a precedence. However we have been working at our personal tempo.”
That’s when the realisation hit: To innovate and scale, NPCI needed to take full possession of its stack. “We can’t be depending on anybody, and whether or not we succeed or fail needs to be our capability and selection,” Kanvaty mentioned.
This determination marked the start of a brand new section—one the place NPCI would construct and management its personal know-how, with no licensed platforms or exterior dependencies. This is identical thought Kanvaty holds for constructing foundational AI fashions in India too.
The Tech Behind the Billion-Individuals Impression
Across the time UPI was being conceptualised, NPCI had already internalised these classes. “We have been assured about it. I might not say that we have been conscious, or knew that it might attain half a billion sometime… However we knew that it was going to be a profitable product,” he mentioned.
Price was all the time a priority, however by no means the only driver. “That’s the motto of NPCI. We now have to provide again to the ecosystem. We’re not for revenue, so we have now to maintain the fee very affordable.”
Open supply was an apparent selection, and UPI’s stack mirrored that. Constructed on Java, the system included in-memory processing with Kafka and Redis. One of the vital defining design ideas was no dependency on conventional databases.
“For us, latency was extraordinarily important,” Kanvaty defined. “We’re only a switching engine… we don’t must have an excessive amount of state… So the database hits will certainly have an issue when it comes to scalability and latency.”
Although they initially used Postgres for particular bindings like cell numbers and machine IDs, most checks now occur in Redis. This determination has enabled UPI to deal with billions of transactions every day with sub-second response occasions.
If Kanvaty have been to construct UPI in 2025 with infinite compute and price range, he mentioned he would nonetheless use the identical structure. Nevertheless, there’s one shift he would think about: changing Java with Rust.
UPI’s democratic design is one in all its most underrated achievements. “One of many elementary issues that we have now completed in UPI is democratise the entire fee service suppliers (PSP) idea,” Kanvaty mentioned. “The friction there’s minimal. I can use my PIN, I can use my identical account quantity, identical debit card. So there’s nothing that I must do bodily.”
Wanting forward, Kanvaty sees AI taking part in a defining position in UPI 2.0. “I feel we may have much more agentic AIs coming in the place quite a lot of help and expertise could possibly be given to customers,” he mentioned.
One clear use case is personalised troubleshooting. “Immediately, if my transaction failed… we give a really generic error saying that the financial institution is down. However now I feel I can go and say possibly the issue is together with your danger financial institution, why don’t you try to use the opposite financial institution that you’ve got?”
AI may additionally optimise gives and playing cards primarily based on context. “Let’s say I’m reserving a ticket, it can provide me much more profit saying that you’ve got this card, why don’t you utilize it to get a greater supply?”
Voice, Vernacular, and Imaginative and prescient
Voice, particularly in regional languages, is one other huge piece of the long run.
“Individuals are used to a kind of Alexa mannequin the place they wish to make funds with out even touching the cellphone,” Kanvaty mentioned. “We now have already began engaged on that; we have now launched voice assistants in at the least two languages and we wish to scale it up.”
On the rising issues about UPI outages, like on Might 12 when the server was down, Kanvaty mentioned, “It’s a scary scenario, so I don’t wish to give any strategies to anybody. I feel individuals ought to proceed to rely upon it, however sure, it’s a studying expertise for us.”
What retains Kanvaty up at night time? “One is resiliency,” he mentioned. “What occurred on the twelfth, we don’t wish to be in that scenario once more. Uptime is important. We perceive the affect UPI has on the nation. The opposite is fraud. These are the 2 huge challenges we have now to remain on prime of and hold bettering continuously.”
In the meantime, NPCI is internally exploring AI instruments like Cursor and others in a managed setting. “We now have created an AI platform for builders to begin experimenting. Nonetheless within the early days, so it could be somewhat extra time earlier than we will say that we’re an excellent adopter.”
Kanvaty is candid concerning the challenges of scaling UPI past smartphones and into rural areas. “We tried numerous choices,” he mentioned. “We tried working with function telephones and voice. Whereas voice-based options are another we’re exploring, they nonetheless don’t supply the sort of expertise that smartphones present.”
The issue isn’t simply technological, it’s about constructing belief and ease of use.
“Extra assist with onboarding is essential,” Kanvaty famous. “We have to work on completely different fashions that can provide rural customers the identical confidence and luxury in making digital funds. Nevertheless it’s going to be an extended and hard journey. There are not any straightforward solutions.”
When requested a few hypothetical future the place UPI could possibly be embedded into wearable gadgets—say, a pendant with tap-to-pay performance—Kanvaty doesn’t dismiss the thought.
In truth, NPCI is already testing adjoining applied sciences.
“We’re exploring palm-based funds, and in addition ‘faucet and pay’ utilizing NFC-enabled gadgets. We even experimented with embedding NFC tags into QR codes, so customers can simply faucet as an alternative of scanning. The early outcomes are promising, with a greater person expertise. However once more, the actual problem is scaling it throughout the nation.”
However What About Fraud?
Amid all of the innovation, a persistent problem looms. The character of UPI fraud has shifted dramatically lately. “Earlier, fraudsters cloned playing cards and replicated chips—it took effort. Now, it’s all social engineering,” Kanvaty mentioned.
“Individuals are tricked into scanning malicious QR codes or getting into their PINs. It’s now not a few fraudster initiating the transaction—it’s the shopper doing it, willingly, on their very own machine,” he mentioned.
To deal with this, NPCI has taken a two-pronged method—consciousness and know-how. “We’ve invested quite a bit in consciousness campaigns,” Kanvaty mentioned, referring to the general public service bulletins that includes actor Pankaj Tripathi. “We tried to make the message catchy, delicate, however efficient.”
On the technical aspect, NPCI has deployed AI fashions that analyse behavioural patterns over a 90-day interval. NPCI assesses the kind of transactions a person has made, and who they’re interacting with, and primarily based on that, the crew can decline doubtlessly fraudulent funds.
“We do about 60 crore transactions every day, and out of those, practically 15 lakh are blocked resulting from suspected fraud,” he mentioned, including that it’s a cat-and-mouse sport as fraudsters finally adapt.
The put up How UPI is Powering a Billion Transactions Each Day appeared first on Analytics India Journal.
