Making useful things,
one thoughtful step at a time.
I'm Kailash Sankar. I've spent 15 years shipping software that has to hold up in the real world.
Notes from the field.
All writing-
Evaluating an MCP Server
Building an evaluation suite for an MCP server that hands a product's data to an AI agent.
-
Defense in Depth: Tenant Isolation for an Agent
Five layers of security designed to prevent cross-tenant data leaks in a code-executing agent.
-
What I Learned Adding Memory to AI Agents
Notes from building agent memory systems with a lot of prompting, directing, and iteration.
Things I've built.
All projects-
systems-labs
A systems engineering playground, and where most of the notes on this site started as experiments.
-
PocketMCP
A local-first MCP server with a web UI, built while the protocol was new enough that writing one was the fastest way to learn it.
-
dev-tools
Small tools for dev and QE teams that run without analytics or data collection, so nothing leaves the machine.
Still curious. Still making.
Current focus
AI agentsEvaluationObservabilityWeb systems
More broadly, I care about systems that are useful, observable, and safe to operate — the kind that stay comprehensible long after launch.
When I'm away from a screen, I'm probably tinkering in my home lab, reading, or out walking Luna.
Open to a conversation