Hey, I'm Saurabh
Data engineer progressing through LLMOps, GPU engineering, and RL infrastructure. I document every layer of the journey — the systems, the tradeoffs, and the things I wish someone had told me.
Each layer builds on the last — from reliable data systems to the frontier of RL infrastructure.
ETL pipelines, feature stores, and the scalable data infrastructure that powers everything above it.
RAG pipelines, vector databases, LLM observability, and serving models reliably in production.
CUDA kernels, distributed training, performance optimization, and building the systems behind reinforcement learning.
I'm building toward RL infrastructure expertise — one layer at a time. Follow along as I document the progression, share what I'm learning, and build in public.
Learn More About MeNotes from the transition.
Every LLM is built on five foundational pillars: Basics, Systems, Scaling Laws, Data, and Alignment. This post maps out what they are and why mastering them is the path to building real AI systems.
Read Article →Test-time training lets models update their own weights during inference. Learn how TTT layers work, their GPU implications, and why this changes AI infrastructure.
Read Article →Learn how to add observability to LLM applications using OpenTelemetry and data engineering practices.
Read Article →Compare ChromaDB, Pinecone, Weaviate, and pgvector to find the right vector database for your RAG application.
Read Article →Learn how to orchestrate a Retrieval-Augmented Generation pipeline using Apache Airflow and data engineering best practices.
Read Article →