Sonny Laskar

Sonny Laskar

Co-Founder

About

Sonny is the Co-founder of Zijus, where he is pioneering the world’s first "PromptCode Full-Stack Agentic" platform. A veteran of the AI landscape, Sonny brings over 15 years of industry expertise across Machine Learning and Product Development.
 
Recognized as one of the prominent technical minds in the region, he has been named multiple times among the Top 10 Data Scientists in India. His relentless drive for innovation is evidenced by his track record as a serial hackathon winner with more than 20 wins.
 
In recognition of his leadership and entrepreneurial contributions, Sonny was recently honoured with the prestigious Young Alumni Achievers Award from IIM Indore. 

We've all talked to voice bots that feel like talking to a walkie-talkie awkward pauses, jarring interruptions, robotic responses. Building a Voice AI that actually feels human isn't just about plugging an LLM into a Text-to-Speech API. It's an exercise in extreme asynchronous engineering.

In this session, we'll walk through what it actually takes to build an enterprise-grade, multi-lingual Voice Agent grounded in real, production architecture rather than theory. Starting from why naive voice pipelines break down in practice, we'll progressively unpack the architectural layers needed to solve real-world problems: acoustic interference, network lag, race conditions between speech and interruption, mid-call provider failures, and live language code-switching.

Using working code from a production Cascade-based voice pipeline as reference, we'll break down concrete strategies for each failure mode what to build, why naive approaches fail, and the specific engineering patterns that get you to consistently low-latency, natural-feeling conversations. We'll also cover where Cascade architectures fit today versus emerging Speech-to-Speech approaches, and what to watch for as that landscape shifts.

You'll leave with a practical architecture checklist and battle-tested strategies you can apply to your own voice AI stack regardless of which framework you build on.

Read More →