Panch Mukesh
I'm a full stack developer with a passion for building scalable systems, exploring cloud-native technologies, and sharing what I learn along the way. On this blog you'll find deep-dives into Python, Kubernetes, API design, developer tooling, and the craft of software engineering — from architecture decisions to everyday productivity tips.
-
LLM Call Traceability: Why You Need It and How to Start with OpenTelemetry
LLM failures return a 200 and a wrong answer. Learn why tracing every call matters, how OpenTelemetry's GenAI conventions model it, and how to get your first trace running.
Read more -
Opus Plans, Sonnet Builds, Haiku Documents. The Hard Part Is the Handoff.
Picking a model per sub-agent takes one line of config. Making the work survive the trip between them takes a handoff contract — here's how to design one.
Read more -
Claude Has Three Different Memories. Here's What Each One Actually Does.
claude.ai, the Claude API, and Claude Code each ship something called "memory" — and they're three unrelated systems. Here's what each one actually stores, and which one you're using right now.
Read more -
Caching While Coding with Claude Code — What's Actually Happening Under the Hood
I tracked 65 days, 1,142 sessions, and 1.83 billion cache read tokens. Here's what's actually happening every time you run Claude Code — and why you're probably already benefiting from it without knowing.
Read more -
How Claude Code Image Paste Actually Works
You see the word "image" appear in the terminal. Here's what's really happening underneath — from keystroke to API call.
Read more -
Behind the Scenes of Your GenAI App
Every time you type a message into an AI-powered chat, a precise sequence of events unfolds in milliseconds. Here's what actually happens — from your keyboard to the model and back.
Read more -
Why Your Tests Pass Locally but Fail in CI
Most developers don't realize that GitHub Actions tests a merge commit — not their branch in isolation. This post explains how actions/checkout creates a temporary merge ref combining your PR with the latest main, why that causes unexpected test failures, and how rebasing regularly keeps your local environment in sync with what CI actually runs.
Read more -
Claude Code Hooks: Auto-Label Every Terminal Session
Replace Claude Code's random session hashes with meaningful labels like S1 and S2, so notifications tell you exactly which terminal needs your attention.
Read more -
Claude Code Hooks: Get Notified When It Needs You
Stop babysitting the terminal. Claude Code hooks let you run scripts on task completion, permission prompts, and idle states — so you get notified the moment Claude needs you, no matter what you're doing.
Read more -
Managing Context Across AI Coding Tools
Context engineering is the meta-skill that separates productive AI-assisted developers from frustrated ones. Learn the universal patterns—project memory, progressive disclosure, and intentional compaction—that work across every AI coding tool.
Read more