At AWAAZ.AI, I own LLM inference and real-time voice infrastructure for a platform handling 200K+ calls per day. I migrated a production Hindi collections agent from managed APIs to self-hosted SGLang inference, cutting daily inference spend by 60% without a regression in agent quality.
I also drove conversational latency from 2.5–3 seconds to 800–900ms and shipped a streaming response architecture that reduced user-perceived latency by 65%. Before AWAAZ.AI, I contributed to AI legal and contract-management products at PENSIEVE and optimized DSP and Dolby audio software at IMAGINATION TECHNOLOGIES.
