Startup Deep Dive : Gnani.ai – from a first profit to a national voice-AI model mandate

In FY25, a Bengaluru voice-AI company that had spent its whole life burning small amounts of money finally posted a net profit of ₹3.19 crore ($3.3 million) on revenue of ₹53.87 crore — a turnaround from a net loss of ₹51 lakh the year before. Months later, the same eight-year-old firm, Gnani.ai, was handed a national mandate few of India’s larger AI labs won: to build a sovereign voice-to-voice foundational model of 14 billion parameters under the IndiaAI Mission.

That is the contradiction worth sitting with. A company with fewer than 220 employees and annual revenue smaller than a mid-sized software services team’s quarterly billing was chosen, from a field of 506 proposals, to help build one of the models the Indian state wants to run on. Gnani.ai did not get there by being the biggest. It got there by being narrow, patient, and stubborn about one problem: making machines understand and speak Indian languages over a phone line.

Quick facts

What they do

Gnani.ai sells voice-first artificial intelligence to enterprises. Instead of a chatbot that types back at you, it builds systems that listen to a caller, understand what they said in their own language and accent, decide what to do, and speak back — the kind of automation a bank uses for loan-collection calls or a lender uses for customer verification. The company describes itself as an “agentic” voice AI platform, meaning the software can carry out multi-step tasks in a conversation rather than reading from a fixed script. Its customers are large regulated businesses, mostly in banking, financial services, insurance (BFSI), telecom, healthcare and automotive.

The origin

Ganesh Gopalan and Ananth Nagaraj met at Texas Instruments. Gopalan had spent more than two decades in technology and marketing roles across Texas Instruments, IBM, Sapient and Satyam, with an MBA from the Indian School of Business; Nagaraj was a signal-processing engineer whose earlier career ran through Texas Instruments, Kyocera and Aricent. In July 2016 they incorporated Gnani Innovations in Bengaluru with a plainly stated goal, in Nagaraj’s words to AI Time Journal, “to make information accessible to large diverse populations.”

The founding insight was specific rather than grand. India speaks dozens of languages, most of its people are more comfortable talking than typing, and most enterprise software assumed the opposite. Gopalan later framed voice AI to Pulse 2.0 as “a large enough unsolved problem that could impact millions.” The pair chose to own the hard part of the stack — the speech recognition engine tuned for Indian languages and accents — rather than wrap someone else’s. That decision, to control the pipeline from streaming audio to a proprietary speech-to-text model, is the thread that runs through everything the company did next.

The struggle years

Building Indian-language speech recognition in 2016 meant building without the luxury Silicon Valley took for granted: abundant, clean training data. English speech models had oceans of transcribed audio; Hindi, Kannada, Tamil and Telugu telephony did not. Gnani had to gather and label its own, one telephone corpus at a time, which is slow, unglamorous work that does not show up in a demo.

The business shape shifted with reality. The early accessibility framing narrowed into something enterprises would pay for: inbound and outbound call automation, letting companies run smarter contact centres. By 2019, three years in, the team was still only about 25 people — speech engineers, NLP specialists and data scientists — a sign of how deliberately, and how leanly, the company grew. Validation came early from Samsung Ventures, which the founders treated as confirmation that their conversational AI actually worked, but scale did not.

The financials show the strain lasted years. As late as FY24 the company was still losing money, posting a net loss of ₹51 lakh. For a firm founded in 2016, that is eight years to the edge of profitability — a long runway on modest capital, and a reminder that deep-tech built the hard way does not compound quickly.

The turning point

Two things happened close together, and they compounded.

The first was commercial. Revenue more than doubled from ₹23.09 crore in FY24 to ₹53.87 crore in FY25, and the ₹51 lakh loss flipped to a ₹3.19 crore profit. The engine behind that was BFSI automation at scale: the company has said its platform helped financial institutions collect over $2 billion from end customers in a six-month stretch around its Series A, including one bank recovering more than $400 million in overdue loan EMIs.

The second was national. On 31 May 2025, IT Minister Ashwini Vaishnaw named Gnani.ai — alongside Soket AI Labs and Gan.ai — as a company selected under the IndiaAI Mission to build an indigenous foundational model, chosen from 506 proposals submitted since the Call for Proposals opened. Gnani’s brief: a 14-billion-parameter, real-time, multilingual voice-to-voice model, initially covering 10 Indian languages. A profitable-by-a-whisker startup had been asked to help build part of the country’s sovereign AI stack. The government mandate did not just validate the technology; it reframed the company from a vendor into national infrastructure.

The money behind it

Gnani has stayed comparatively capital-light for a deep-tech firm, raising roughly $21.9 million across about seven rounds since 2017 (as per Tracxn). The shape of the funding:

• Early backing (2017 onwards): Samsung Ventures came in as an early investor, which the founders cite as validation of the core speech technology.
• Series A — July 2024: ₹30 crore (about $4 million) led by Info Edge Ventures, with Samsung Ventures already on the cap table. Earmarked for sales growth and expansion across geographies (as per Entrackr and Mobility Outlook).
• Series B — 2026: a round led by Aavishkaar Capital, headlined at $10 million with participation from existing backer Info Edge Ventures; regulatory filings for the first close show ₹68 crore (about $7.17 million) (as per Entrackr and BusinessToday).

What each backer changed: Samsung Ventures gave the company early technical credibility; Info Edge Ventures (the Naukri parent’s fund) provided the growth capital that took Gnani from a niche vendor to 100-plus enterprise customers across India and the US; Aavishkaar Capital, an impact-focused investor, is funding the global expansion and the agentic AI push. The 2026 round reportedly lifted the valuation to ₹818 crore ($87 million), up from a reported ₹204 crore earlier — roughly a fourfold jump between rounds.

How it makes money

Gnani earns by automating conversations that enterprises would otherwise staff with people. The mechanics:

• Enterprise deployments: banks, lenders, insurers and telcos pay to run voice and multichannel agents for collections, verification, support and outbound outreach.
• Owning the pipeline: because Gnani controls its own speech-to-text and text-to-speech models rather than licensing them, more of each contract’s value stays in-house — the margin sits in the model, not in reselling someone else’s API.
• Products that anchor contracts: the Inya.ai platform lets enterprises build no-code conversational agents across voice, chat, email and SMS; Armour365 sells voice biometrics for identity verification; proprietary small language models are tuned per sector.
• Value framed as ROI: the company markets outcomes — it cites up to 70% cost reduction and 80% first-contact resolution, and points to driving $20 million in revenue for an automotive OEM as a reference deal.

The part people get wrong: this is not a generic “AI chatbot” reselling a large foundation model. The defensible asset is a decade of proprietary Indian-language telephony data and models trained on it — company-stated at 14 million hours of telephonic audio across 40-plus languages — which is expensive to gather and hard for a rival to replicate quickly.

The numbers

Gnani’s reported financials (unit: ₹ crore) show a company that stayed small and near break-even for years, then inflected sharply in FY25. Figures are as per regulatory filings reported by Entrackr; conversions use $1 ≈ ₹96.0 as of 18 September 2026 (Trading Economics).

• FY25 revenue: ₹53.87 crore, up roughly 133% from FY24’s ₹23.09 crore (Entrackr, regulatory filings).
• FY25 net profit: ₹3.19 crore, the first profit in the company’s history, reversing a ₹51 lakh loss in FY24.
• Headcount: about 213 employees as of May 2026, up from roughly 25 in 2019 (Tracxn; AI Time Journal).
• Patents: 10-plus deep-tech patents filed, per company and press accounts.

A caveat on precision: some databases list a slightly higher FY25 revenue figure (around ₹56.95 crore), likely reflecting total income including other income rather than operating revenue. The ₹53.87 crore operating-revenue figure is used here because it appears consistently across reporting of the MCA filings.

Where the money comes from

The revenue base is concentrated, by design, in regulated Indian enterprises with high call volumes:

• BFSI is the core: named clients include TVS Credit, the Bajaj Group, Fibe and Muthoot Finance (as per Inc42). Collections and verification are the highest-volume, highest-value use cases.
• Scale of usage: the platform processes more than 30 million voice interactions daily across 12-plus languages, serving over 200 enterprise clients (company-stated, widely reported).
• Geography: primarily India, with a growing US presence — the company reported crossing 100 enterprise customers across India and the US around its 2024 Series A, and is using the 2026 round to widen international sales.

The surprise is where the future revenue may come from. The IndiaAI Mission mandate points Gnani at sectors it did not historically bill — education, healthcare and agriculture — including a pilot in rural Uttar Pradesh using voice AI to give health guidance to expecting mothers. That is a different customer (often the state) and a different economics from BFSI collections.

The risks

• Customer and sector concentration: revenue leans heavily on BFSI, and much of it on collections. A regulatory clampdown on automated collection calling, or a downturn in lending, would hit the highest-margin use case directly. When a small number of large contracts drive growth, losing one resets the year.
• Competition from foundation-model giants: global players with far deeper pockets are shipping multilingual speech models. Gnani’s moat is its proprietary Indian-language telephony data, but if general-purpose models close the accuracy gap on Indian languages, the “own the pipeline” advantage narrows. The company is small — ₹53.87 crore of revenue — against opponents spending that much in a week.
• Execution risk on the sovereign model: building a 14-billion-parameter voice-to-voice model is a research undertaking on a national deadline and shared GPU pool. Delays, cost overruns or accuracy shortfalls would be public and reputational, not just internal — the flip side of being chosen as infrastructure.

The takeaway

Gnani.ai’s story argues for depth over breadth. The founders picked one genuinely hard problem — Indian-language voice over telephony — and spent eight years and only about $22 million owning it end to end, rather than chasing a wider, shallower product on more capital. That patience looked like slowness for years: 25 people in 2019, a loss as recently as FY24. Then the same narrowness became the reason a government picked them to build national infrastructure, and the reason the numbers finally turned. The transferable lesson is that a defensible data asset, built slowly and unglamorously, can be worth more than speed — but only if you survive long enough to reach the moment the market, or the state, needs exactly what you spent years hoarding.

Frequently asked questions

What does Gnani.ai do?

Gnani.ai builds enterprise voice AI: speech-to-text and text-to-speech models tuned for Indian languages, an agentic conversational platform called Inya.ai, and voice biometrics under Armour365. Banks, insurers, lenders and telcos use it to automate calls for collections, verification and customer support.

Who founded Gnani.ai and when?

It was founded in July 2016 in Bengaluru by Ganesh Gopalan (CEO) and Ananth Nagaraj (CTO), former colleagues at Texas Instruments. The legal entity is Gnani Innovations Private Limited.

Is Gnani.ai profitable?

Yes, as of FY25. The company reported a net profit of ₹3.19 crore on revenue of ₹53.87 crore in FY25, its first profit, reversing a net loss of ₹51 lakh in FY24 (as per regulatory filings reported by Entrackr).

Why was Gnani.ai selected by the IndiaAI Mission?

On 31 May 2025, the government named Gnani.ai — with Soket AI Labs and Gan.ai — from 506 proposals to build an indigenous foundational model. Gnani’s task is a 14-billion-parameter, real-time, multilingual voice-to-voice model covering Indian languages.

How much has Gnani.ai raised and what is it valued at?

Gnani has raised roughly $21.9 million across about seven rounds, including a ₹30 crore Series A (July 2024, Info Edge Ventures) and a 2026 Series B led by Aavishkaar Capital. The 2026 round reportedly valued it at ₹818 crore (about $87 million).

Sources

Figures are as of September 2026. Currency converted at $1 ≈ ₹96.0 as of 18 September 2026 (Trading Economics).

• Entrackr — Gnani.ai FY25 financials and Series B round led by Aavishkaar Capital (2026)
• Entrackr — Gnani.ai raises $4 million Series A from Info Edge Ventures (July 2024)
• BusinessToday — Gnani.ai raises $10 million from Aavishkaar Capital (March 2026)
• Aavishkaar Capital — Series B announcement and valuation (2026)
• Mobility Outlook / ANI — Gnani.ai ₹30 crore Series A details (July 2024)
• Inc42 — IndiaAI Mission selects Soket AI, Gnani.ai and Gan.ai (May 2025)
• BW Businessworld / The Week — IndiaAI foundational model selection and GPU pool (May 2025)
• NewsBytes / IndiaAI.gov.in — Gnani’s voice-to-voice model, languages and rural pilot (2025)
• Tracxn — funding total, rounds, headcount and company profile (2026)
• Tofler / ZaubaCorp / InstaFinancials — Gnani Innovations Private Limited, CIN and incorporation (MCA data)
• Pulse 2.0 — interview with CEO Ganesh Gopalan on products and business model
• Business Standard — Gnani.ai voice-first AI, products and traction (July 2024)
• AI Time Journal — interview with co-founder Ananth Nagaraj on origins and early team (2019)

Found an error? Write to us and we’ll correct it in the open, dated, on the piece.