This blog is a collection of those small steps. Consider this post epoch 0: the weights are initialized, and training starts here.

Who's writing

I'm Gayatri, co-founder of Nuvolink, where we work on agentic systems, machine learning and inference. On paper, that's my title. In practice, I'm a prototyper and builder: I'm happiest taking a rough idea and turning it into something that actually runs. I borrowed that title from a post by Boris Cherny on where engineering work is heading, and it describes what I do better than any job title I've had.

Why I'm writing

Most of what I learn while building ends up scattered across notebooks, half-finished docs and terminal history I'll never open again. The insights are real, but they disappear.

So I'm writing them down. Partly for myself, because writing forces clarity: if I can't explain why something worked, I probably don't understand it yet. And partly for you, because some of the most useful things I've read came from people honestly sharing what they tried, what broke and what they figured out.

What you'll find here

• Agentic systems: how to design, build and debug multi-agent workflows, and where they quietly fail

• Evals: how to tell whether an AI system actually works, beyond "the demo looked good"

• Inference engineering: making models faster, cheaper and practical to run

• Search and retrieval: what makes RAG systems better or worse in real domains

• ML and DL: fundamentals, models and ideas worth understanding properly

Some posts will be hands-on experiments. Some will be literature surveys that map out a topic. Some will be short notes on an idea, a mistake, or something I read and couldn't stop thinking about.

How I think about problems

The part of building I enjoy most happens before any code: defining the problem, taking it apart, and finding the core issue underneath the obvious one. A lot of these posts will walk through that process, not just the final answer.

I also like crossing domains. I started out building line-following and maze-solving robots, moved through anomaly detection systems and survival models, and now spend most of my time on AI systems for legal research and other real-world applications. Each domain teaches you something the others don't, and I'll bring those connections in where they help.

About the logo

The ∇ (nabla) is the symbol for a gradient. The orange dot beneath it is where every descent is headed: the minimum, or at least somewhere slightly less wrong.

Come along

If you're building in this space, or just curious about how these systems work under the hood, follow along. And if you're working on similar problems, I'd love to hear from you.

Here's to the first step, and to being slightly less wrong with each one.