The archive · AI & Models · Product decision · 2024–2026
Uplift AI bets voice-first AI, not text, onboards regional-language users
Uplift AI (YC S25) bet voice, not text, onboards non-reading billions: Urdu-first models earned a $3.5M seed and 2,500 Khan Academy videos.
Uplift AI
What the business is
Uplift AI builds speech and text-to-speech models plus APIs for underserved languages, starting with Urdu, Sindhi and Balochi, so developers can add voice interfaces to apps, e-commerce, banking, tutoring and public services.
Starting capital：$3.5 million seed (announced 2026-01-29) led by Y Combinator and Indus Valley Capital, with Pioneer Fund, Conjunction, Moment Ventures and Silicon Valley angels participating.
How it started
After leaving Apple, co-founder Hammad Malik spent a year reading two decades of voice-technology research and concluded smaller languages were an opportunity a fast small team could win before big tech moved. What began as a side project to build datasets for translation and voice became Uplift AI when the WhatsApp bot experiment showed how many users voice-only could reach.
What happened
The team publicly launched text-to-speech APIs for Urdu, Sindhi and Balochi on Hacker News on 2025-08-18, reporting that Khan Academy was already dubbing videos to Urdu. Because transcription and labeling tools are broken for these languages, Uplift built its own data-gathering, labeling, training and voice infrastructure instead of buying data. In January 2026 it announced a $3.5M seed led by YC and Indus Valley Capital, and YC's later launch post described Orator, a 32M-parameter Urdu voice model trained on character voices recorded in studios in Karachi, Lahore, Quetta and Islamabad and labeled by 50 contractors.
How it ended up
The company is still active as of 2026-09-05: Y Combinator's directory lists Uplift AI as an active Summer 2025 company with a three-person team in San Francisco, and its stated scope has widened from Pakistani languages to regional languages globally.
Background
Uplift AI, founded in 2024 and part of Y Combinator's Summer 2025 batch, builds speech recognition, text-to-speech and conversational voice models for languages big tech serves poorly, starting with Urdu, Sindhi and Balochi. Its pitch is that text interfaces exclude billions of people who cannot read, and voice is the on-ramp that lets them use digital services.
The company began as a side project to make datasets for translation and voice models. A hacked-together Urdu voice bot on WhatsApp, given to one domestic worker, reached 800 users in two days, convincing the founders, including ex-Apple engineer Hammad Malik, that voice-first access was a real market rather than a research curiosity.
Because transcription models, spell correctors and labeling tools barely exist for these languages, Uplift chose to gather its own data, build its own labeling tools and train and deploy its own models rather than buy data. By its public launch on 2025-08-18 it offered text-to-speech APIs and reported Khan Academy was already dubbing educational videos into Urdu.
In January 2026 the company announced a $3.5M seed round led by Y Combinator and Indus Valley Capital. As of 2026-09-05 it remains active with a three-person team in San Francisco and is expanding its model coverage beyond Pakistani languages.
What has to be true
- Big tech under-invests in languages whose monetizable markets look small today, leaving room for a fast specialist team to ship better models first.
- The WhatsApp bot experiment was cheap, direct proof of demand from non-reading users, validating voice-first before any serious capital was spent.
- Sourcing and labeling data in-house produced the right data distributions, which the founders argue yields better models with far less data than rivals that buy datasets.
- Selling infrastructure APIs rather than one consumer app let partners like Khan Academy and Syngenta put voice in front of education and agriculture users, spreading adoption beyond a single market.
- Recording authentic district accents and everyday personas for Orator turned a model into cultural credibility, differentiating Uplift from generic synthetic voices.
What can be applied
Incumbents under-invest in under-served languages, so a tiny team that owns data, labeling and training can win; one humble WhatsApp pilot proved demand before serious capital.
Aftermath
As of 2026-09-05, Uplift AI is an active YC Summer 2025 company with a three-person team in San Francisco. Its $3.5M January 2026 seed, led by Y Combinator and Indus Valley Capital, funds expansion of regional-language voice AI; CWPakistan reported more than 1,000 developers on its APIs, 2,500 Urdu videos published by Khan Academy and Syngenta using its tools for farmer guidance. YC's page describes foundational voice models for regional languages, sold through simple APIs and SDKs, to support voice e-commerce, banking and tutoring.
Sources
- Launch HN: Uplift (YC S25) – Voice models for under-served languages
- Uplift AI Secures $3.5 Million Seed Funding To Expand Voice AI For Regional Languages
- Uplift AI: Foundational Voice Models for regional languages
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card