The full story.

I’ve spent a decade watching AI projects break. Almost never at the model.

Over the last decade, I’ve worked across the full spectrum of Analytics and AI — from collecting and engineering raw data, to building machine learning systems, to deploying them in production and on resource-constrained edge hardware.

I’ve never been interested in staying within a single discipline. If a project needed data engineering, I learned data engineering. If it needed forecasting, I built forecasting systems. If it required embedded AI, I went down to the silicon. If deployment became the bottleneck, I focused on deployment.

The hardest part of AI is rarely the model. It’s the system around it.

That mindset has taken me through a deliberately diverse set of problems: real-time telecom fraud detection, industrial IoT platforms, demand forecasting, Snowflake-native AI applications, embedded face and liveness detection, radar-based human sensing, and Agentic AI systems for enterprise workflows.

Today, as Chief AI Scientist at Brillersys, I build production AI systems while mentoring engineers, working with leadership teams, and bridging the gap between research ideas and real-world deployment.

The more I work across domains, the clearer one pattern becomes: AI is not a separate discipline. It’s the next layer of automation — built on good data, sound engineering, and thoughtful system design.

That’s also why I write.

Through articles, talks, tutorials, and The MG Signal, I share practical lessons from building AI in production — without hype, without shortcuts, and without pretending every problem needs the latest model.

If you’re building AI that needs to work outside a notebook, you’re in the right place.