A layered approach to agent evaluation: from deterministic checks and trajectory evals to LLM-as-judge and human review.

Three things that broke in production and exactly how I fixed them

Why we stopped routing agents one step at a time and compiled the full pipeline upfront

How splitting conversation and understanding into two changed everything

Why a single database eventually stops being enough, how sharding splits data across multiple nodes, and why consistent hashing is the elegant solution to the painful problem of adding or removing shards.

What CQRS is and why it separates reads from writes, what Event Sourcing means when you store events instead of state, how they fit together, and the real tradeoffs before you reach for either pattern.

Why Kafka partitions become unequal, what happens when one partition hogs all the work, and the concrete strategies teams use to fix hot keys, rebalance load, and prevent consumer starvation.

Things that bit me building production agents with LangGraph

A relaxed Uttarakhand travel story filled with train journeys, hill roads, temples, lakes, and a lot more walking than planned.

An introduction to this section where I'll be sharing travel stories, personal reflections, and meaningful life experiences.