What I build
AI products, agent systems, and the infrastructure beneath them.
I like work where software, product thinking, and cloud architecture have to agree with each other instead of taking turns pretending to matter most.
Varun Sai · AI engineer · software developer
I build AI systems that survive the demo.
I work across cloud infrastructure, software, agent workflows, and the product decisions around them. VarunSai.in is where I keep the useful parts visible: what I built, what I learned, what broke, and what became worth writing down. Over time, this grows into notes, letters, project rooms, and a clean starting point for people building something serious with AI.
Cloud-first
Agent workflows
Human in the loop
Tea before hype
Not just a homepage. The beginning of a library, a build log, a letter archive, and a better way to show how the work actually gets made.
Approach
A home for the work, not just the announcement.
The interesting part of AI engineering is rarely the screenshot. It is the structure behind it: the choices, the constraints, the debug trail, the cloud decisions, and the taste required to keep the whole thing human on the other side. That is the layer I want to publish here.
What lives here
This starts simple, but it is not meant to stay small.
Live now
/notes
Working Notes
Working notes, architecture fragments, references, and half-formed ideas worth keeping close.
Live now
/builds
Build Log
Projects, tools, and systems that made it past the first sketch and earned a cleaner explanation.
Warming up
/newsletter
Letters
Longer writing on AI engineering, product judgment, cloud decisions, and the people around the tools.
Warming up
/debug
Debug Archive
Postmortems, repair notes, and the causes behind fixes that should stay visible.
What I believe
The model is not the whole product.
Reliability, interface, guardrails, handoffs, writing, and judgment matter just as much as the intelligence inside the system. Sometimes more.
What stays personal
Good taste is part of the build.
So is honesty. Tea still outranks coffee, and if something is not solved yet, it should not be dressed up like it is.
Signals
Some work I keep nearby while building my own.
Reference shelf
lil-agents
Ryan Stephen’s lil-agents is the kind of repo I like to keep open: small, readable, and not trying too hard to look important. It is a good reminder that agent systems can stay simple and still be sharp.
Direction
What I want this site to become
A growing body of work around notes, builds, letters, and debug trails, with enough clarity that someone new can understand both the result and the reasoning behind it.
Collaborate
If you are building with AI, I can help give it shape.
Sometimes the first version needs a builder. Sometimes the hard part is deciding what the workflow should be, where the product boundary belongs, or when a human should stay in the loop. I like that layer of the work, and I like taking it all the way to a solid rollout.
Where I can step in
- Turn a rough AI idea into a buildable scope, workflow, or product direction.
- Prototype or ship agent workflows, internal tools, and AI product features.
- Help with cloud setup, deployment paths, cleanup, iteration, and the messy middle.
How I like to work
Collaboratively, honestly, and all the way through. I prefer useful first versions over inflated promises, and I stay interested after launch when the real behavior starts showing up.
From here
A calm starting point. A larger body of work on the way.
For now, it is a homepage. Soon, it becomes a sharper archive of what I build, what I learn, and who I work with.