Uyir Lab
AI that talks the way people actually talk.
Big models learn from the written web. People speak in dialects, slang and two languages at once. We build models that understand them, starting with Tamil, and put them to work where being understood matters most.
Our mission
We train AI on real human voices.
Most AI learns from the written web, then guesses how people talk. We start with the people. Native speakers record how they really speak, at home, with family, in two languages at once, and that is what our models learn from.
- 01
Real people, real speech
Voice recordings and conversations from native speakers in their own dialects. Not text written by another AI.
- 02
Given with consent
Every speaker chooses to take part and knows how their recordings are used. Nothing is taken without asking.
- 03
Checked by a person
A reviewer listens to every recording before it is used to train a model.
- 04
Trained from scratch
Nila, our first model, is built from the ground up on Tamil, English and Tanglish, so it learns how people talk from the start.
Our bigger vision
Every language, and every way of speaking it, should have a life in AI.
Uyir means life in Tamil. Today AI works best for people who speak standard English. We are building a lab that gives each underserved language its own model, built with the people who speak it, so anyone can talk to AI, and to the services they rely on, in the language they think in.
See the languages we're working on- 1language live todayTamil
- 1language up nextSinhala
- 14languages on our roadmapIncluding Punjabi, Swahili, Hausa, Amharic and more
Our approach
Big labs go wide. We go deep.
One language at a time, we collect real conversations from native speakers, with their consent, and turn them into an understanding of how the language is really used. Depth beats breadth when your users are the ones being misunderstood.
Products
One way of understanding people, three ways to use it.
Every product is built on the same research: our own Tamil model, native-speaker tests and speech people chose to give us, so each language we add helps all of them.
- Pilots open
01 · For cities
Townline
AI that helps run City Hall.
AI for permits, 311, public records, licensing and inspections. Residents talk or type in English, Tamil or both; it checks the city's rules, files into the city's own systems, and keeps the city in control of its records, rules and tests.
- Answers from pages the city approves
- Files into the city's own systems
- Staff portal with corrections and tests
- Pilots open
02 · For any team that takes calls
Voice agent
Talks like the people who call.
A real-time voice agent for support lines, clinics, banks and telecoms. Callers speak English, Tamil or both mixed, and it answers in their language, then logs the call in English for your team.
- Understands dialects and code-switching
- Answers by voice or chat
- Logs every call in English
- Early access
03 · For AI companies
Language Brain
Make your model understand how people really talk.
Our models trained from scratch, and native-speaker evaluation sets, for AI labs and teams building agents. Plug in the dialects, slang and code-switching your model gets wrong today.
- Language Brain API
- Models trained from scratch, starting with Nila
- Native-speaker reviewed evals
Also from the labNila, our Tamil model →Uyir Bench report card →Give your voice →
Give your voice
Teach AI how you really talk.
Record a few sentences in your own language and dialect, the way you speak at home. Every recording is given with consent, checked by a person, and helps the next model understand people like you.
Recording is open in Tamil today, in any dialect. Other languages open as we reach them.
- 1Pick your dialect
- 2Read or speak freely
- 3We review it, then Nila learns
From the lab
Nila is training right now.
Nila 1B, our own model with a billion parameters, trained from scratch on 9.6 billion tokens of Tamil, English and Tanglish (25 billion tokens of training in all). These numbers come straight from the training run.
nila · live
Connecting to the training run…
Training loss
Languages
The languages big models miss.
We start where the gap is widest: languages with tens of millions of speakers and very little of the data big models learn from.
- தமிழ்Live
Tamil · 86M+ speakers
Spoken Tamil differs sharply from written Tamil, and dialects from Jaffna to Chennai barely appear online.
Visit Tamil → - සිංහලNext
Sinhala · 20M+ speakers
Written and spoken Sinhala are almost two different registers, and models trained on formal text sound stiff in conversation.
Join the waitlist → - پنجابی·ਪੰਜਾਬੀPlanned
Punjabi · 156M+ speakers
About 89M Punjabi speakers in Pakistan (2023 census) write in Shahmukhi, a Perso-Arabic script, but much of the language lives in speech, not text. Many speakers write online in Urdu or English instead.
Join the waitlist → - KiswahiliPlanned
Swahili · 97M+ speakers
Most of its speakers use it as a second language, and everyday speech mixes in English (Sheng in Nairobi), which formal web text misses.
Join the waitlist → - Harshen HausaPlanned
Hausa · 94M+ speakers
One of Africa's most spoken languages, yet with little web text, written in both Latin and Arabic (Ajami) script.
Join the waitlist → - አማርኛPlanned
Amharic · 60M+ speakers
Its own script (Ge'ez) and rich word forms get split into many pieces by tokenizers built for English.
Join the waitlist → - Basa JawaPlanned
Javanese · 68M+ speakers
Speech levels such as ngoko and krama change the words depending on who you are talking to, and many speakers write online in Indonesian instead.
Join the waitlist → - FulfuldePlanned
Fula · 39M+ speakers
Spoken across West and Central Africa and written in Latin, Arabic (Ajami) and its own Adlam script. Common Crawl's language detector does not even recognize it.
Join the waitlist → - Èdè YorùbáPlanned
Yoruba · 50M+ speakers
A tonal language where meaning rides on diacritics that most web text leaves out.
Join the waitlist → - भोजपुरीPlanned
Bhojpuri · 52M+ speakers
Mostly spoken, often written in Devanagari mixed with Hindi, and not recognized as a separate language by Common Crawl's language detector.
Join the waitlist → - Afaan OromooPlanned
Oromo · 45M+ speakers
45M+ native speakers, yet fewer than 1 in every 100,000 web pages is in Oromo.
Join the waitlist → - پښتوPlanned
Pashto · 55M+ speakers
Strong regional dialects and very little high-quality digital text.
Join the waitlist → - Asụsụ IgboPlanned
Igbo · 34M+ speakers
Tonal, with many dialects, and the little text online is often missing its tone marks.
Join the waitlist → - ភាសាខ្មែរPlanned
Khmer · 21M+ speakers
Written without spaces between words, so standard tokenizers and search struggle from the first step.
Join the waitlist → - Kreyòl ayisyenPlanned
Haitian Creole · 13M+ speakers
Spoken by nearly everyone in Haiti, while French dominates formal writing, so little text reflects how people talk.
Join the waitlist → - Runa SimiPlanned
Quechua · 7M+ speakers
A family of mostly oral varieties with little standardized text to learn from.
Join the waitlist →
Work with us
Bring Uyir Lab to the people you serve.
Businesses, governments, community organizations and teams building agents: tell us what you need.