களம் – Kalam — Episode 01 · Intro

நன்றி for stopping by — welcome to களம் (Kalam), the field where Tamil.AI gets built in public, episode by episode.

I'm building Tamil.AI: it started as a Tamil-learning app for diaspora kids, but the real goal underneath it is deeper — real Tamil speech technology, starting with a problem generic AI still gets wrong: ழ், the retroflex sound that trips up transliteration, TTS, and every "supports Tamil" claim you've seen. From there: the dialects too — Kongu, Madras Bashai, Madurai Tamil — none of which have a real training corpus today.

களம் is where that work happens on camera: the models, the data collection, the dead ends, the decisions — not a polished highlight reel, the actual field notes. Every conversation recorded here also becomes part of an open corpus, so the series builds an audience and a dataset at the same time.

This first episode is the intro — who's building this, why Tamil, and what களம் covers from here.

If you care about Tamil tech, dialect representation, or just want to watch something get built in the open, this is for you.

ழ Tamil.AI — Tamil Speech Technology · Build in Public