Model Progress Update: What Users Told Us, and What We Need Next | களம் · Kalam — Tamil AI

We said we would build Tamil AI in public. This episode shows what that means in practice: real users tried the pronunciation feature and told us what they actually want.

Model Progress Update — two things from this episode:

1. We need more voice samples.
Our pronunciation scoring is only as good as the voices behind it. So far it has been tested on a small first batch of speakers. To make it work for more people, we need a wider mix of voices: more speakers, more ages, more regions of Tamil Nadu and the Tamil diaspora.

2. Users asked for a demonstration video for every letter.
After trying the app, one viewer shared that it was interesting, and asked whether we could add a visual picture showing how to position the mouth to pronounce each Tamil letter. That is exactly the feedback we hoped for. So we are planning a short demonstration video for each letter.

Why this matters: build in public only works if people join in. This feedback came from someone using the product, not from us guessing. It tells us what to build next.