Hello!

I've been in the room.

Not the room where someone explains AI with a neat diagram and a TED talk voice. The room where the model just degraded in production. Where the VP wants answers in 20 minutes. Where three teams have three different definitions of "done" and none of them are right. Where the governance framework doesn't exist yet because nobody thought to build one before the launch. That room.

I'm Amrita Thavrani - Senior Technical Program Manager at Amazon, where I lead AI platform programs that operate at a scale.

In the last few years alone, I've launched Amazon's first enterprise prompt engineering platform, driven AI model coverage from 87% to 95%, pushed classification accuracy to 99% on the company's highest-traffic paths, and led the global compliance modernization that delivered $3.2B in annualized cost savings. Before Amazon, I've shipped programs across Citigroup, Infosys, IBM, and Oracle - across continents and industries.

Why this exists

AI moves fast. The jargon moves faster. And most explainers are written by people who've read about AI - not people who've been accountable for it.

I kept noticing the same thing: brilliant people - engineers, product managers, executives, analysts - nodding along in meetings and quietly Googling what it meant afterwards. Not because they weren't smart. Because nobody had explained it plainly, without the hand-waving, without assuming you already knew what RAG stood for or why your LLM keeps hallucinating on edge cases.

So I started writing it down. The stuff I wish someone had handed me on day one. The frameworks I built from scratch. The mistakes I've made so you don't have to make them first.

Less hype, more signal. Field notes from a TPM who's shipped it.

What you'll find here

The Explainers - clear, jargon-free breakdowns of AI concepts, tools, and terminology. Written for people who need to understand it, use it, or talk about it without sounding like they're reading from a glossary.

From the Field - practitioner-level posts drawn from real programs. What it actually looks like to build an LLM governance framework. Why prompt engineering is a program management problem. What I'd do differently if I were launching an enterprise AI platform today.

If you've ever sat in a meeting, heard a term you didn't recognize, smiled confidently, and then frantically searched for it - this site is for you.

Hope you enjoy reading it as much as I enjoy curating it.

Say hello: thavrani.amrita@gmail.com Connect: linkedin.com/in/amritathavrani