I'm Kumar Mukesh, and I build AI products.

15 years of making large technology programs land for global financial institutions, 6 of them onsite in the USA as the client-facing lead for BNY Mellon, New York Life and The Hartford. The work was governance frameworks and delivery assurance, plus the recovery jobs nobody volunteers for.

These days I am a Manager at State Street's Charles River Development, accountable for delivery and production assurance on an Azure SaaS platform carrying $58T+ in assets. The biggest team I have run came earlier, at BNY Mellon: 30+ direct and indirect reportees across the USA, India and Mexico. Different clients, same job. Work out what is actually breaking, put a cadence around it, then automate whatever should never have been manual in the first place.

The last few years I have pointed those same instincts at AI. Asset Assistant came out of that, a search tool that finds a file by what is inside it rather than what it was named. I also model delivery risk in Python, where the interesting part is rare, expensive events hiding inside a lot of routine signal. An MSc in Data Science is nearly finished, on top of a PG Diploma from IIIT Bangalore.

I write about the whole of it: building with AI, working out whether it actually works, and the ideas sitting underneath the tools. One thing does keep resurfacing. Most teams I have watched ship without agreeing what "working" means, then argue about quality afterwards with no shared evidence to argue from. That is a governance problem in a machine learning costume, which is why I keep ending up in the middle of it.