Uplift AI is teaching voice software Urdu, Sindhi and Balochi

Hammad Malik's first test user was his driver. Malik had spent about nine years building voice technology at Apple and Amazon, and he put together a simple Urdu WhatsApp bot and handed it over to see how it fared. Dawn reported that within two days 800 people were using it. That small experiment is the origin of Uplift AI, a voice AI company that builds speech models for Urdu, Sindhi, Balochi and, next, other Pakistani languages, and sells them to developers and businesses through an API.

The company is a bet that the languages big technology firms skip are worth building for, even when the customers have little money. In late January 2026 it announced a $3.5m seed round. The numbers are small by Silicon Valley standards. The ambition is not.

Two ex Big Tech engineers and an Urdu data problem

Uplift AI was founded in March 2024, according to Dawn, by Malik and Zaid Qureshi. Malik is chief executive. His profile on the Y Combinator directory describes roughly nine years on Apple's Siri and Amazon's Alexa teams and a Cornell degree, and a separate profile in WASSSL adds that he studied at LUMS. Qureshi, the chief technology officer, previously designed AWS Bedrock Guardrails, according to the same directory. An August 2025 WASSSL profile also lists a third founder, Muhammad Bin Sabir, as chief operating officer, though the funding coverage names only two.

The first money was small. Malik told Dawn that friends gave him $5,000 each and that he raised about $125,000 in November. The company later joined Y Combinator's Summer 2025 batch, where the directory listed a team of three based in San Francisco. The company's own website now gives its base as San Francisco and Lahore, and its open jobs on the Y Combinator page were for engineers in Lahore, at monthly pay between $1,500 and $2,500 plus a sliver of equity.

Its flagship model, Orator, is described in the Y Combinator profile as a very small 32 million parameter system inspired by a published speech architecture known as VITS2. It was trained on voices recorded in studios in Karachi, Lahore, Quetta and Islamabad, with about 50 contractors hired to label the data. That is a modest model by the standards of the large language labs, and that is rather the point. The scarce input is not compute but clean recordings of people speaking.

How Uplift AI sells Urdu text to speech and AI calling

The business is business to business. Uplift AI's site offers a text to speech studio, a developer API and AI calling, where an agent phones customers and speaks Urdu. It lists more than 100 voices with different accents, characters and styles, and prices them on a ladder: a free tier, then plans at $5, $50 and $300 a month, then custom enterprise contracts with on premises deployment. The $50 Pro plan, for example, includes 25 hours of speech generation and 1,250 minutes of calling.

The voice agent product is priced in rupees. Its documentation puts conversation time at 12 rupees a minute, with calls placed from Pakistani phone numbers over infrastructure hosted in Pakistan to keep latency down. The use cases it lists are the unglamorous ones that fill call centres: confirming cash on delivery orders, rescheduling deliveries, booking appointments, chasing payments. Dawn reported that Uplift AI uses OpenAI and other providers only for the language model layer and builds its own voice infrastructure for local languages. In plain terms, it rents the reasoning and owns the voice.

Khan Academy, Syngenta and a $3.5m seed round

The round, reported by Dawn on 30 January 2026, came from Y Combinator and Indus Valley Capital, with Pioneer Fund, Conjunction, Moment Ventures and Silicon Valley angels also named. Coverage disagrees on who led. Techlist says Y Combinator did, TechJuice says the round was backed by Y Combinator and Indus Valley Capital, Dawn names both without saying who led, and i2i Ventures' later funding review says it was led by Indus Valley Capital and Y Combinator. Dawn reported that more than $1m of the money would be spent in Pakistan on generating original voice data. Techlist put the figure at about $1m for data collection and labelling and said the company expects it to create thousands of local jobs. That is a forecast, not a headcount.

Customers are few and named. Khan Academy used the technology to dub educational videos into Urdu. The count has moved: WASSSL reported 104 videos in August 2025 with thousands more planned, while the January announcement said 2,500. The agrochemical firm Syngenta is building voice assistance for Pakistani farmers on the models. Dawn noted that Syngenta employs nearly 1,000 people whose main job is to answer farmers' voice queries, and that the quality of responses varies widely, which is the gap Malik says voice AI can fill. TechJuice reported more than 1,000 developers building on the API.

“We want to be the best in the world for languages no one else wants to build for”

The hard parts: cheap customers, scarce data and big rivals

Malik was candid with Dawn about the economics. Voice AI remains expensive even in developed markets, he said, and deploying it where purchasing power is a fraction of that in the West is harder still. He also noted that only around 6pc of Y Combinator companies become massive successes. The data problem is real as well. Fewer than 5pc of Pakistanis speak Balochi and even fewer can write it, leaving almost no usable data online, so Uplift AI had to build datasets from scratch.

Then there is competition. WASSSL's profile claims Google, OpenAI and ElevenLabs have struggled with Urdu, and that Uplift AI's models cut dubbing time tenfold. Both statements are the company's claims and no independent benchmark was published in the coverage we reviewed. The strategic risk is the obvious one: if Pakistani languages prove commercially valuable, larger labs can pay for their own recordings. Uplift AI's defence is to get there first and to hold the data.

The global pitch is the part that makes the Pakistan story larger. Malik told Dawn he wants to supply the language layer to hardware and platform companies, including robotics firms and virtual reality makers. His reasoning, in his words: "If someone speaks Pashto, they would want to speak in Pashto to their robots." TechJuice reported that the roadmap also includes broader language coverage, especially in the MENA region.

What comes next for Uplift AI

Dawn reported in January that Punjabi, Pashto and Saraiki were due later in 2026, with Punjabi first, alongside the existing Urdu, Sindhi and Balochi. We found no public announcement of those launches in the coverage reviewed for this article, so the roadmap should be read as a promise. The company has not published revenue, and the latest public financial figure remains the January 2026 seed round.

The test for Uplift AI is whether paying customers in Pakistan can fund a voice stack that the rest of the world might one day want. It is a plausible bet, and an early one.

Reported from Dawn, Techlist, TechJuice, WASSSL and i2i Ventures, the Y Combinator directory and Uplift AI's own website and documentation.