Skip to content

About

Hi, I'm Kislay Kumar, a Data/AI engineer at Lyric Tech. Last six years have been backend and data engineering: distributed systems, pipelines, large-scale data processing. The AI part is more recent. I build with LLMs day to day, mostly LangChain and LangGraph, and I kept hitting the same problem. I couldn't get things done without really understanding why they worked.

So I pick something I use, or something I've read about and only half followed, take it apart properly, and write down what's actually going on. The posts are the notes I wanted to find.

What I work with

LLM systems in production: retrieval-augmented generation, multi-agent workflows, tool integration and evaluation. Underneath that, the backend and data work that keeps them standing up.

LLM systems

  • LangChain
  • LangGraph
  • Hugging Face
  • Python

Data and backend

  • Kafka
  • Spark
  • Iceberg
  • Debezium
  • Postgres
  • MongoDB
  • Kubernetes
  • Docker

Before Lyric I was at Hevo Data, working on data integration, CDC pipelines and the move to microservices, and before that Microsoft. Computer science at BITS Pilani.

How the posts work

Every equation gets three things: the formula, what each symbol means in plain words, and what would go wrong if you removed that term. The third is the one most explainers skip, and it's where the understanding actually lives.

Intuition comes before formalism. You should feel the problem before you see the solution.

And when I don't understand why something works, the post says so. I'd rather write that than fake it.

The series

How We Got Here walks from 2012 to now. Every post answers one question: what was broken, and what fixed it.

Research

Separately, under Research: a question, a decision rule committed before any test data is scored, and whatever the rule returned. Including the times it returned no.

If a post made something click, I'd like to hear it. If I got something wrong, I'd like to hear that more.