Skip to content
Omkar Shendge

Omkar Shendge

I build distributed systems in Go and run them on bare metal until they break. This is where I write down what broke and why.
Start with the News Feed Read an incident GitHub

Three ways in

Section titled “Three ways in”
Building a news feed in Go Thirteen posts, in build order — schema and recursive CTEs, RabbitMQ fan-out, idempotency keys, the transactional outbox, then pgvector search and an eval harness that refused to flatter the results.
Incidents Things that broke in production on my own hardware, written up honestly — including the wrong hypothesis I chased first, because that part is usually the interesting one.
Distributed systems from scratch MIT 6.5840 labs in Go — MapReduce, then a key/value server with a distributed lock. Written to understand the primitives rather than to use them.
Platform & homelab A two-node Kubernetes cluster on a 2008 desktop and a Ryzen box. GitOps with ArgoCD, self-hosted object storage, backups with a tested restore.

What this is

Section titled “What this is”

Every post here comes from something I actually built and ran, usually on hardware that gave me a reason to think harder — a control plane with 4 GB of RAM, a CPU too old for modern container images, a GPU that lost driver support years ago.

Constraints like that are useful. They turn “which database should I use” into a question with a real answer, and they surface failure modes that a managed platform would have quietly absorbed on my behalf.

Currently

Finishing a real-time chat system in Go — WebSockets with Redis fan-out, Kafka as the durable log, pgvector for semantic search over history, and a RAG assistant running on a local model. Next up: WebRTC calling that works between two real networks, and load-testing the whole thing until something gives.