The same problem we solved for India, now needed solving in Arabic.
When we built Akshara and Ragini, the insight wasn’t technical, it was linguistic. People don’t speak in the clean, formal register that most voice AI is trained on. They mix languages mid-sentence, carry regional accents, and expect a system that keeps up. That held true across Hindi, Tamil, Telugu, Kannada, and six other Indian languages. Turns out, it’s just as true in Arabic.
Introducing Alif and Rawi
Alif is our bilingual Arabic-English speech-to-text model, built to handle the real, messy, code-switched conversations happening in MENA contact centers every day, not the formal Arabic that most commercial ASR systems are trained on.
Rawi is our native Arabic text-to-speech model, engineered to speak with the natural rhythm, intonation, and multi-dialect flexibility that actual enterprise voice interactions demand.
Together, they’re ConvoZen’s first step into West Asia, and they were built from the ground up for how Arabic is actually spoken, not adapted from an existing model.
Why this matters
Akshara and Ragini are now deployed across 40+ enterprises in nine Indian languages, including HDFC Bank, Tata AIG, Apollo, Maruti Suzuki, and Cars24. What we learned building and scaling those models shaped how we approached Alif and Rawi. Same philosophy, new geography.
The MENA expansion is the next logical step for a platform built on one core belief: conversational AI only works when it’s built for the people actually using it.
Read the full story here.