Hari Kharel

Notes on building with AI agents: architecture, context, evals and cost. The parts the demos skip.

Start reading Browse topics
Scroll

Latest

Agents

What I'll be writing about

TL;DR This blog is about building with AI agents after the demo works.

· 1 min read Read →

All writing

Topics

What I write about.

Agents

Orchestration, multi-agent swarms and keeping them observable.

Context engineering

Knowledge graphs, context packs and what to put in the window.

MCP

MCP servers and Apps that put internal tools inside Claude and ChatGPT.

Evals & testing

Deterministic checks for non-deterministic systems.

Memory

Lesson stores and shared memory so agents improve between runs.

Platforms & guardrails

Sandboxes, templates and safe defaults for AI-built software.

Practice

Interview prep, built in the open.

Study

Learn the pattern, not the question.

One visual page per concept interviews test: step-through animations and runnable code in Python, TypeScript and C#.

Start studying → Sandbox

Then solve it under interview conditions.

Staged problems with hidden tests, a step debugger and an AI interviewer that hints but never gives the answer. Sign-in required.

Open the sandbox →

About

I'm Hari, a product engineer in London. I build AI platforms and agents, and write here about what works, what breaks and why.

What I believe about building AI CV Study Sandbox LinkedIn Email