15+ Indian languages
Speech recognition built on Sarvam's Saaras models, trained for Indian speech — Hindi, Tamil, Telugu, Bengali, Marathi, and more, not English with an accent filter.
Case study · In-house product
Desi Caption
AI captions for Indian-language video.
The problem
01 Captioning tools weren't built for how India speaks.
Captions decide whether a video travels — most viewers watch on mute, and platforms rank on watch time. But mainstream captioning tools are English-first: feed them Hindi, Tamil, or the Hinglish that creators actually speak, and you get gibberish to clean up by hand, minute by minute.
Desi Caption starts from Indian speech instead. Upload a video, and accurate captions come back in the creator's own language — with Hinglish transliteration for the way audiences actually read — ready to export and post.
What we built
02 Upload, transcribe, export. That's it.
Hinglish, handled
Code-switching is how India actually talks. Transcripts come with Hinglish transliteration so captions read the way the audience speaks.
Word-level timing
Every word carries its own timestamp, so captions land in sync and editors can cut to the syllable.
SRT & VTT export
One click exports standard subtitle files that drop straight into YouTube, Instagram, Premiere, or any player.
Uploads that scale
Videos go straight from the browser to object storage via presigned URLs — a 500 MB file never touches the API, and only ~1 MB of extracted audio goes to transcription.
A pipeline that recovers
Transcription runs as queued background jobs with retries, so a flaky network or a huge file never takes the product down — the transcript just appears when it's done.
Building for the Indian market?
We build products end to end — the same way we build our own.