01
Correct a translation
The fastest, most useful thing you can do. Every correction on the translator is reviewed and folded into the corpus that trains the next model. Ten minutes of your time outlives you.
Open the translator ↗Take part
The corpus is built by speakers, not by us.
There is no dataset to download and no expert to defer to. What Kodava takk will look like in a machine's memory is being decided right now, by whoever shows up to correct it.
01
Ways to help
02
Add words we're missing
Household words, farming words, ritual words, words your grandmother used that nobody says any more. Especially the ones with no English equivalent.
Submit a word ↗03
Lend your voice
Kodava has almost no recorded audio. Native pronunciation recordings are the raw material for speech tools that don't exist yet — and the only way a great-grandchild hears the accent.
Write to us →04
Mentor someone
An hour a month, against a real admissions deadline, is worth more to a seventeen-year-old in Gonikoppal than any donation. Connect is looking for its first mentors now.
About Connect →05
Tell us what you check before you sell
If you grow coffee or pepper, the most useful thing you can give us is an hour of honesty about what you actually look at, and what you argue about on the phone. Market is being designed around those conversations.
About Market →06
Publish your work
Musicians, dancers, poets, film-makers, photographers. Arivāme gives Kodava creators a stage that isn't buried in someone else's feed.
About arivāme →07
Moderate and review
A community platform for a small community lives or dies on native speakers who care. If you can judge whether a translation is right, we need you.
Write to us →08
Tell your people
Kootas, family groups, the WhatsApp thread that never sleeps. The diaspora is scattered, and word of mouth is still how it finds anything.
Become an ambassador →02
What happens to it
Where your correction ends up.
It gets reviewed by a native speaker. If it holds up, it goes into the verified pairs that the translator draws on — so the next person asking something similar gets your answer, not the machine’s guess.
Those verified pairs accumulate into a parallel corpus: the dataset that does not exist for Kodava anywhere in the world. At enough scale it becomes training data for a dedicated Kodava model that runs without renting anyone’s API.
Which means a correction typed on a phone in a coffee estate or a flat in Bangalore is, quite literally, a line in the first machine that will ever speak Kodava properly.
- 01You correctA translation you know is wrong
- 02A speaker reviewsNative-speaker moderation
- 03It shipsInto the live translator
- 04It accumulatesInto the parallel corpus
- 05It trainsThe dedicated Kodava model