Vinod Kumar

I'm a Staff Software Engineer at ServiceNow, where I work on software systems at scale. I'm an NIT Meghalaya graduate, and I've been writing production code for 12+ years — APIs, distributed systems, platform infrastructure, the kind of work where uptime isn't optional.

A few years ago, I started going deep on AI — not the research kind, but the engineering kind. How do you actually wire an LLM into a production system? What breaks when you do? Why does RAG work beautifully in demos and fall apart on real corpora? How do you make an agent reliable instead of impressively unreliable?

I didn't find a lot of good writing on those questions. What I found was either too theoretical (papers), too shallow (tutorials), or too opinionated without the workings shown. So I started writing it myself — first in private notes, then here.

Every article on NeuralSys is the kind of thing I needed when I was figuring something out. I write from first principles, with real examples, and I try to be honest about what I don't know. That's the deal.

NeuralSys is a technical publication on AI systems engineering. It covers agents, context engineering, RAG, knowledge graphs, inference, and the architecture decisions behind production AI. The writing is aimed at software engineers who are serious about building with AI — not just prompting it.

The philosophy: content first. Excellent writing, excellent reading, excellent engineering. No hype, no clickbait, no "Top 10 AI tools" lists. Just careful technical thinking, written for engineers who want to actually understand what they're building.

The name? Neural networks meet systems engineering — that's the intersection I live in.