i built an llm that recommends fixes for slow database queries in a week. this is the story of the months of evals, validators and cluster probes it took to make its advice honest.

a friendly, agent-assisted field guide to the staff engineer transition, where the job becomes less about solving every problem yourself and more about turning ambiguity into durable motion.

an in progress dump from trying to turn ai coding speed into runtime proof instead of a bigger pile of plausible green checks.

what started breaking once rocksdb crossed from useful embedded store into a scale-sensitive production subsystem.

the harness could finally build something, but the first real simulators taught me that better prompts do not fix a visual system with too much freedom.

a growing set of small claude code automations, each tied to a specific chore i already do. a living list i'll keep adding to as i build more.

in-progress notes from building a jira-to-pr agent for apache pinot, the over engineering instinct i started with, the signals that told me when to add each layer and the things i'm still figuring out.

i spent a year hand-building visual explainers in cursor. then i tried to automate it with ai. the first month was all agent infrastructure and zero visuals.

a real incident walkthrough for when your system is healthy and completely wrong

a practical explanation of snapshots, delta lake, apache iceberg and why data teams suddenly care about table formats.

the art of removing costs you didn't know you were paying on every row.

what it took to make a stable diffusion pipeline fast enough to replace twitter profile pictures with anime versions as you scroll.

achieve harmony in complex clusters using finite-state machines

unleashing the full potential of your rocksdb with the right configuration

before diving into this article, i should state that — as a developer — any situation requiring placing a language a project into a language b package should occur very rarely. most of the time it’s preferable to consider re-designing the interaction between various language components in these situations. but what if this situation is unavoidable? open source projects such as apache flink and apache spark serve as examples. these projects have been written completely in java but also have python modules available for those who don’t want to use the java api.

groovy functions

one of the primary advantages of using pinot is its pluggable architecture. the plugins make it easy to add support for any third-party system which can be an execution framework, a filesystem, or input format.

flood

the lab

leader and followers

a glimpse at capabilities which makes airflow better than its predecessors

explore the various strategies to deploy ml models in apache flink/spark or other realtime data streaming applications.