Eelam Lab builds and owns the models it trains, including Nila. Contributors keep ownership of their original recordings and writing; you grant us a licence to use them for training. We intend to give the Tamil community meaningful access to what we build.

Contributions are reviewed, cleaned and used to train Nila language and speech models. Models or datasets may be licensed to third parties, as described in the consent you give. We do not sell your contact details.

What about my privacy?

We collect only what we need: your name, email, country, dialect and interests. Data is stored encrypted with access limited to our team. We follow Canada's PIPEDA principles. Read the Privacy Policy for detail.

How do I delete my data?

Email privacy@eelamlab.com from the address you signed up with. We will confirm and delete your contact details within 30 days. Contributions already used in a trained model cannot be removed from that model, but will be excluded from future training.

Do I need to speak "proper" Tamil?

No. We want Tamil exactly as you speak it: your dialect, your slang, your mix of Tamil and English. That is the whole point.

Eelam Lab · Building Nila, a Tamil language model

Teaching AI to speak Tamil, the way Tamils speak it.

யாதும் ஊரே யாவரும் கேளிர்

Every place is home, everyone is kin. · Kaniyan Pungundranar, Purananuru 192

Contribute your voice Read the manifesto

அ
ஆ
இ
ஈ
உ
ஊ
எ
ஏ
ஐ
ஒ
ஓ
ஔ

Manifesto

Tamil has been spoken, sung and carved into stone for more than two thousand years. It is one of the oldest living classical languages on earth. Yet in the data that teaches machines to speak, it is barely a whisper. Eelam Lab is an independent lab building Nila, a Tamil language model trained from scratch, by and for the Tamil community. Not translated. Not an afterthought. Trained on how we actually speak.

years of continuous literature: 2,000+
speakers worldwide: 80M+
of typical web training data: <0.1%

A living timeline

Two millennia of Tamil, in one scroll.

c. 300 BCE - 300 CE

Sangam literature

Over two thousand poems of love and war, mapped onto five landscapes: mountain, forest, field, sea and desert. Poetry as an atlas of feeling.

c. 1st century BCE - 5th century CE

Thirukkural

Thiruvalluvar's 1,330 couplets on virtue, wealth and love. Seven words each, four then three, translated into dozens of languages.

c. 850 - 1279 CE

Chola inscriptions

Temple walls became public records. Land grants, village assemblies and royal deeds were carved in Tamil across stone, from Thanjavur outward.

Centuries, into the 1800s

Palm leaf manuscripts

Knowledge was incised on dried palm leaves and copied by hand, generation after generation. Many classics survived only because someone kept copying.

1578

The printed word

Movable Tamil type arrived, and with it newspapers, textbooks and a reading public. Scholars raced to rescue palm leaf texts into print.

1931 onward

Cinema and music

Film songs, radio and the stage carried spoken Tamil into every home, and gave the language new rhythms, slang and stars.

1800s - today

The diaspora

From Jaffna to Toronto, Madurai to Kuala Lumpur, Tamil now lives on every continent, in dozens of dialects and code-switched with many languages.

Now

Artificial intelligence

The next chapter is being written by machines trained mostly on other languages. We intend to write it in Tamil, with Tamils.

What we are collecting

Every way Tamil lives.

A model only knows what it has heard. We are gathering Tamil as it is really used: at home, at the market, on stage and on the phone.

How it works

From your voice to the model.

Sound becomes script becomes tokens.

Impact so far

Built thread by thread.

⚑ Placeholder figures. Live numbers arrive at launch.

Hours of audio: 1240
Contributors: 860
Dialects covered: 9

The kolam

Every contribution adds a thread to the pattern.

A kolam begins as a grid of dots and becomes one continuous line looping around them. A language model is the same: thousands of separate voices, joined into one pattern no single person could draw alone.

13 threads

Contribute

Lend your Tamil to the future.

Join the contributor list, then record or upload straight away in the contribution studio.

Already signed up? Open the studio →

From the journal

Notes on teaching AI to speak Tamil.

All articles →

Questions

Asked and answered.

Eelam Lab builds and owns the models it trains, including Nila. Contributors keep ownership of their original recordings and writing; you grant us a licence to use them for training. We intend to give the Tamil community meaningful access to what we build.