Ex-Air Force Strategist's AI Startup Covers 2,331 Languages

ZWAG AI's Linguistics Intelligence Platform now covers 2,331 languages, with plans to push beyond 5,000 — and the startup behind it just closed its pre-seed round after a founder who failed three times in five years finally got investors to say yes.

Priya M. Nair, founder of ZWAG AI and co-creator of the Mātṛ model with Nipuna Abeykoon, spent her early career in the UAE Air Force & Air Defence, then 11 years in strategy and growth at an American engineering and programme management company. A 2019 mini-stroke put her at high medical risk. The pandemic that followed forced the pivot.

"While millions lost their lives, I have been granted the opportunity to live, so what next," Nair told Gulf News, recalling the period between 2020 and 2021. "What do I want to be my legacy, how do I want to be remembered."

She chose entrepreneurship as the vehicle for skills she already had: leadership, strategy, communications and creativity. The hard part wasn't the product — it was the money.

What nearly killed the startup?

Fundraising, by Nair's own account, concentrated almost every challenge she faced. Much of the work went into convincing the market and investors of two things: that a non-technical founder leading a deep-tech company is not a liability, and that neither is being a woman.

The pre-seed round closed recently, and Nair is precise about what unlocked it. "What made the difference was referral and networking, trust built steadily over years," she said.

The record behind that trust is stark: "In five years of entrepreneurship I failed three times before Mātṛ. That is what it took to raise our round and set a pathway to success."

How does Mātṛ fix linguistic bias?

Most large AI models train predominantly on English and a handful of widely represented languages. The result: they struggle with complex grammar, cultural context and linguistic nuance.

Mātṛ takes a different technical route. Instead of forcing languages into standardised patterns, the model mathematically identifies where languages diverge and corrects AI interpretations at inference time — without requiring existing models to be retrained.

Two examples show the stakes. In Japanese, standard models push indirect, respectful honorifics through English-centric logic, so a polite request comes out jarringly blunt; Mātṛ calculates the divergence and restores the tone, intent and cultural context the speaker actually meant.

In legal discovery, errors cut deeper. In languages with complex case marking, such as Turkish or Malayalam, conventional models flatten passive and conditional statements into direct assertions. Nair's example: a witness's cautious "It could have been known" becomes a definitive "They knew." Mātṛ preserves that evidentiary nuance.

The market case is scale, not sentiment. As healthcare, education and public services adopt AI, linguistic accuracy decides whether communities get included in the digital future or left out of it. Language, as Nair puts it, represents identity, knowledge and access.

Why the UAE?

Nair argues the UAE is among the most linguistically diverse societies in the world, and that few places bring together as many languages and cultures in a single society. "That diversity is an asset the rest of the world cannot easily replicate," she said.

Her proposal: a Centre of Excellence for linguistic sovereignty based in the UAE — a place where native speakers and native translators from many languages work together on the data and the judgements AI systems depend upon. Her reasoning is blunt: "Linguistic accuracy cannot be built from English-language assumptions alone. It requires the people who hold the language."

Such a centre would not serve the region alone, in her view. It would contribute to democratising AI globally, with the work carrying the UAE's name. "In the Middle East, I cannot think of anywhere else it could realistically be built," she said.

What else is on the runway?

Nair also mentors founders through Startup with Priya and wrote the book Formula-G: Scale Your Startup to a Growth Machine. Her advice to aspiring women entrepreneurs is operational: develop deep expertise, question established assumptions, think globally from the outset — and arrive with far greater reserves of preparation, optimism, patience and resilience than male counterparts.

With pre-seed capital in the bank and a 5,000-language roadmap, Mātṛ's next test is whether inference-time correction can scale into the enterprise stacks that need it.