Real-time voice AI for customer conversations, handling text-to-speech, speech-to-text, and multilingual support.
Resolved booking workflow failures, implemented regex-based phone number extraction directly from live call transcripts, and removed unnecessary number pronunciation confirmation steps for 100% automated booking execution.
Completed core functional voice agent updates and integrated high-naturalness Indian English TTS voices for field deployment across hotel and clinic call centers.
Replaced simple Q&A pipeline with an end-to-end reasoning loop (User speaks → STT → LLM reasons → Tool call → Execution → Response) enabling real-time room bookings and appointment scheduling.
Optimized total roundtrip latency budget by tuning model selection, trade-offs between model size and Time-to-First-Token (TTFT), achieving sub-second response times on standard CPU servers.
Adding four additional Indian regional languages to STT and TTS.
Real-time vector search retrieval for live support agents during active call flows.