21d • 🏆 Wins
Currently building voice infrastructure for a matrimony company in Karnataka.
Interesting constraint: most of their inbound is in Kannada.
Standard English-language voice agents are useless here — the moment a customer hears an accent that doesn't match their context, they hang up.
Two weeks of testing went into getting the Kannada TTS to not sound robotic.
Switched from ElevenLabs to Sarvam. Iterated on the prompt structure 9 times.
Added Kannada filler words ("hmm", "haan") to make pauses feel natural.
The technical work nobody talks about: voice agents in vernacular languages are 10x harder than English-only deployments.
Every word has to feel native. One robotic syllable and the call is dead.
But the market opportunity is also 10x bigger.
600M+ people in India transact in vernacular languages. Almost nobody is building serious voice infrastructure for them.
Reminder for anyone building: the languages, markets, and demographics that get ignored by global products are usually the highest-margin opportunities for local operators.
3
3 comments
Currently building voice infrastructure for a matrimony company in Karnataka.
skool.com/learn-ai
Start & Scale your own AI Automation Agency. Learn to build AI solutions, land clients & get paid. From the creators of the AAA business model.
Leaderboard (30-day)
1
+473
2
+320
3
+305
4
+238
5
+217
Powered by