I've spent the last two years pushing AI agents into the messy reality
of production - where hallucinations cost money, memory
matters, and "it worked in the demo" doesn't cut it.
Today I run an AI customer service SaaS called DiscutAI. Handling 15,000 messages a day.
That's where I learned what actually breaks - and how to build
things that don't.
Before AI agents, I shipped ML models, built Modern Data Stack platforms from scratch, and earned multiples AWS certification.
I also run a Data Engineering unit at an engineering school - because teaching forces you to understand what you actually know.
Most AI projects fail because someone forgot there's a production after the demo. I didn't. Having worked end-to-end — infra, data, models, agents — I've learned where things tend to bite in production.
We deploy AI agents that handle customer conversations end-to-end — where customers already are. Not a widget. Not a chatbot. They check orders, send emails, book meetings — they don't just reply, they resolve.