Partner with us

Register

Sign In

Register now

Praveen
Gunasekaran
Sr. Director & Chief AI Architect,Engineer
Visa
My journey with artificial intelligence began at 18 when I first heard about AI. I realized AI can take a system from automatic to autonomous. I immersed myself learning AI or whatever courses that led me to it like Cybernetics and Steganography, laying the groundwork for what would become a lifelong passion. From building algorithmic trading systems on Wall Street to developing the Generative AI platform at Visa that’s now driving enterprise-wide adoption, I’ve spent 16+ years at the intersection of engineering and AI innovation. I’ve led teams that reduced ML deployment times from months to weeks, architected systems processing billions of data points daily, processing trillions of pixels in an hour, and pioneering trusted , secure and safe Multi AI agentic systems to achieve 10x productivity gains across global operations. What drives me isn’t just the technical challenge but ensuring equitable access. Having witnessed how technological monopolies can create lasting inequality, I’m committed to democratizing AI. This technology should uplift humanity broadly not concentrate power in the hands of a few. I believe we’re at the dawn of a new scientific revolution, one that will allow us to solve problems beyond our current imagination. My work focuses on building the bridges that will ensure this revolution benefits everyone, creating a future where our children can focus on even greater challenges.
Button
29 May 2025 09:45 - 10:30
Tackling hallucinations, drift & data decay at scale
LLMs don’t fail overnight, they degrade quietly. From prompt fragility to distributional drift, maintaining reliability in production demands more than one-time testing. This session explores how engineering teams are: → Detecting and mitigating drift across prompts, embeddings, and outputs → Designing feedback loops that surface hallucinations in real-world use You'll leave with proven strategies to reduce failure rates, preserve model accuracy, and ensure your LLMs stay robust as data, users, and use cases evolve.

Useful links
link
Useful links
made with Acara
Social media