Eelam Lab · Journal

Notes on teaching AI to speak Tamil.

How we are building Nila, a Tamil large language model trained from scratch: dialects, speech data, tokenizers, culture and the people who make it possible.

• Recording TipsSpeech DataHow to Record a Good Voice Sample on Your PhoneHow to record a good voice sample at home: choosing a quiet room, placing your phone, avoiding noise and echo, and what to say when contributing Tamil speech.October 6, 2026 · 7 min
• Tamil DiasporaHeritage LanguageTamil Diaspora: Keeping a Heritage Language AliveHow the Tamil diaspora in Canada, the UK, Malaysia and Singapore keeps Tamil alive, why heritage speakers matter, and why diaspora voices belong in Tamil AI.September 15, 2026 · 7 min
• Oral HistoryFamilyTamil Oral History: How to Record Your Elders' StoriesA practical guide to recording Tamil oral history with your parents and grandparents: what to ask, how to set up, consent, and how to preserve their stories.August 25, 2026 · 7 min
• ThirukkuralTamil LiteratureThirukkural and AI: What an Ancient Text TeachesWhat the Thirukkural, Tamil's celebrated book of couplets, teaches us about language data, careful transmission, accuracy and building AI for Tamil.August 4, 2026 · 7 min
• TanglishCode-SwitchingTanglish: Tamil-English Code-Switching and AITanglish is how many Tamil speakers really talk. Learn how Tamil-English code-switching works, why AI struggles with it and why models must handle it.July 14, 2026 · 7 min
• Speech DataTamil AITamil Speech Dataset: Why Voice Data MattersWhat makes a good Tamil speech dataset, why today's voice AI still mishears Tamil, and how your recorded voice can help build better Tamil technology.June 2, 2026 · 7 min
• Spoken TamilDiglossiaSpoken Tamil vs Written Tamil: Understanding DiglossiaSpoken Tamil vs written Tamil explained: what diglossia is, how centamil differs from everyday speech, real examples, and why it matters for learners and AI.May 12, 2026 · 6 min
• Tamil AITokenizationTamil Tokenizer Explained: How AI Reads Tamil TextHow does a Tamil tokenizer turn text into tokens? Grapheme clusters, Unicode, BPE and why English-first tokenizers waste Tamil, explained with examples.April 21, 2026 · 8 min
• Jaffna TamilTamil DialectsJaffna Tamil vs Indian Tamil: Key Differences ExplainedJaffna Tamil vs Indian Tamil: how vocabulary, verb endings, pronunciation and loanwords differ, with real examples, and why both matter equally for Tamil AI.March 31, 2026 · 6 min
• Tamil DialectsTamil LanguageTamil Dialects: A Guide to How Tamil Is Spoken WorldwideA plain-language guide to Tamil dialects, from Jaffna and Batticaloa to Chennai, Madurai, Kongu, Tirunelveli, Malaysia, Singapore and the diaspora.February 24, 2026 · 7 min
• Tamil AILow-Resource LanguagesAI Tamil Language Support: Why AI Gets Tamil WrongWhy is AI Tamil language support still so weak? Low-resource data, agglutination, script, diglossia and dialects explained, plus what can fix it.February 3, 2026 · 7 min
• Tamil AILarge Language ModelsTamil Large Language Model: Building One From ScratchWhat does it take to build a Tamil large language model from scratch? Data, tokenizers, dialects, consent and evaluation, explained in plain language.January 14, 2026 · 8 min