A homegrown AI startup is quietly rewriting how Delhi's small traders, clinics, and logistics firms handle everything from customer calls to tax filings — and the window to get in early is closing fast.
Sarvam AI crossed 10 million API calls per month in June 2026, according to figures the company shared with investors last week. That number matters less for its size than for what it represents: a made-in-India large language model, trained heavily on Hindi, Hinglish, and seven other Indian languages, is now processing more real business transactions in Delhi's commercial corridors than most people in those corridors realise.
The timing is not accidental. With the central government's India AI Mission having disbursed roughly ₹2,000 crore in compute grants to domestic AI developers since January 2025, the infrastructure for local-language AI finally exists at a price point that a Karol Bagh garment wholesaler or a Lajpat Nagar clinic can actually afford. Monthly access to Sarvam's voice API starts at around ₹3,500 — less than most businesses spend on a part-time data-entry hire.
The core product is a voice agent that handles inbound calls in Hindi or Hinglini without the robotic lag that plagued earlier IVR systems. A pharmacy on Rajouri Garden's main market stretch, for instance, can route prescription refill requests, check stock availability, and confirm home delivery slots — all without a human picking up the phone during the 11 a.m. to 2 p.m. rush. Sarvam claims response latency under 800 milliseconds, which is close enough to human conversation that callers rarely notice the difference.
Beyond voice, the company's document-processing module has found traction with the dense cluster of chartered accountant firms in Nehru Place, Delhi's long-standing IT and financial services hub. The tool can parse GST invoices, cross-reference them against a client's tally data, and flag discrepancies before a filing deadline. Several CA practices there reportedly cut invoice-reconciliation time by 60 percent in a pilot that ran through March and April 2026.
Delhi's logistics sector — the fleets of delivery operators working out of warehouses along NH-48 and the Kundli-Manesar-Palwal Expressway — is another early adopter cohort. Route optimisation is not new, but Sarvam's integration with existing WhatsApp Business accounts means a fleet manager can receive Hindi-language route alerts and driver updates through a channel they already use, rather than buying new hardware or retraining staff on a foreign-language dashboard.
Sarvam is not without rivals. Google's Gemini has added stronger Hindi support in its 2.5 Pro release, and Microsoft Azure's AI services are aggressively courting enterprise clients through its Aerocity office campus near Indira Gandhi International Airport. Both come with deeper integration tools and stronger uptime guarantees. For a mid-size company — say, a 200-employee textile exporter in Okhla Industrial Area — those enterprise-grade assurances matter.
What Sarvam has that neither of those giants can easily replicate is cultural granularity. The model was fine-tuned on Indian legal documents, government scheme descriptions, and regional commercial language patterns. When a kirana store owner in Pitampura asks the voice agent about GST exemption thresholds for his product category, it answers in the right register, with the right terminology, rather than producing technically accurate but contextually jarring English-translated output.
The company is running a Delhi-specific onboarding programme through July and August 2026, partnering with the Confederation of All India Traders, which has offices in Chandni Chowk. Businesses that sign up before August 31 get three months of free API usage and a dedicated setup call in Hindi. The CAIT tie-up is significant: the confederation claims over 80 million trader members nationwide, and its endorsement signals that Sarvam is prioritising the mass-market small-business user rather than chasing large-ticket enterprise contracts first.
For Delhi business owners still on the fence, the practical advice is straightforward: run a pilot on one narrow use case before July ends. Inbound call handling or invoice parsing are low-risk entry points. The cost of doing nothing is rising faster than the cost of a three-month trial — competitor firms in your category are already testing this, and the learning curve, while modest, does take a few weeks to climb.
How does this story make you feel?
Spread the word
About this article
Published by The Daily Delhi
Daily brief
Free, in your inbox before 7am. Weekdays.
More in
