I'm Mayank — a solution architect who builds AI systems and cares about being able to show they work. My route here was sideways: electronics, then software engineering, then systems architecture, picking up machine learning along the way. AI systems are systems first, and the questions that decide whether one succeeds — how it fails, what it costs, how you measure it, who trusts it — are architecture questions in new clothes.
What I keep coming back to is the evidence layer. Anyone can build something that demos well; the craft is in the measurement — metrics whose assumptions match the problem, splits that don't lie to you, and being able to say precisely how well a system performs and where its limits are. Most of what I write here is about that, and about bringing the same rigour to agentic systems.
Elsewhere in my head
Decision science and probability — and the places different fields turn out to be describing the same problem. Reasoning well under uncertainty is genuinely hard. Most of us are poorly calibrated by default, base rates are easy to skip past, and a well-calibrated range tends to feel less useful than a confident wrong number. I find that difficulty more interesting than any single technique for getting around it.
Get in touch
Email me at mayank.alok@outlook.com. I'm always glad to hear about evaluation problems that turned out harder than expected, cases where a simpler method beat a sophisticated one, or work on agent evaluation and data governance. I read everything and reply to most.
New posts by RSS — I write when I have something worth writing, not on a schedule.