@codoyevskyy
Account based inSouth Asia!
About this account
- Account based in
- South Asia
- Connected via
- South Asia Android App
X says this location may be affected by a proxy or VPN.
Account-level information from X, not a live location or the device used for a specific post.
yak shaving my way into DevOps & distributed systems getting nerd-sniped by ML inference on the side
20 | pune
Joined June 2025
- Tweets7.4K
- Following866
- Followers2K
- Likes33.7K
Pinned Tweet
So I made my portfolio as a Kubernetes cluster.
Yes, it's not just hosted on Kubernetes, the portfolio itself is a cluster
I wanted to make something different from the usual portfolio where you have a hero section, some skills, projects and a contact form. So instead of just mentioning Kubernetes as one of my skills, I decided to make the entire portfolio around it.
Every section of the portfolio is actually running as a separate workload in the cluster. Different namespaces, services and Pods, with the requests being routed through Traefik. The projects and blog sections even fetch live data from GitHub and DEV. to instead of just having static content sitting there.
The homepage is basically a live dashboard of the cluster. It queries the Kubernetes API every few seconds and draws the actual state of the cluster, Pods, nodes, namespaces, events, CPU and memory metrics. There's a 3D topology graph where you can click on a Pod and see its live resource usage, or switch to a 2D table view if you want the data without the spectacle. The sidebar shows which Pod is serving you right now, with a pulsing status dot next to each nav item.
So if a Pod goes down, gets restarted or you (me) scale something, the portfolio actually reflects it. You can literally see the infrastructure behind the website while using the website.
I also added the other stuff you'd expect from an actual deployment- TLS, monitoring, NetworkPolicies, rate limiting, CI/CD and a small Go service that talks to the Kubernetes API.
I built the whole thing locally with k3d and then deployed it on an AWS EC2 instance using the same setup as i didn't wanted to use EKS just for this.
The main idea was more towards my infra skills being the proof. The infrastructure isn't something running behind the portfolio. The infrastructure is the portfolio.
link:- faizn.site
Finally set up remote SSH access to my college PC using Tailscale, so I can remotely access its RTX 4000, 20GB VRAM GPU for my inference engineering experiments.
Now I just need the machine to stay online 24/7 and it won't be down as already it has “DO NOT SHUT DOWN” note on it
Looking to connect with people into Inference Engineering!
I’m loving this rabbit hole and working on implementing Speculative Decoding on my college’s RTX 4000 (20GB VRAM) GPU as part of a college project.
and best thing about it is that these experts aren't explicitly programmed to specialize in coding, mathematics, or reasoning. their specializations can emerge during training as the model learns which experts are useful for different token representations. (4/5)
A girl from my class died of dengue. We weren't close, she was just someone who existed around me. I thought I'd talk to her someday, or maybe I didn't even care if I did. But now she's no more.
It's such a strange feeling realising death is inevitable and unpredictable as well
Many of you don't know this, but I lead the AWS Student Builder Group at my college, where we host hands-on tech sessions.
This time, we’re going online.
ft. @unifiedhadron join us for a hands-on workshop on building an AWS network from scratch.
Registration link in comments.
register here meetup.com/aws-sbg-i2it/even…
Faizan retweeted
I'll be conducting a hands on session on AWS networking at the International Institute of Information Technology, Pune.
about me, worked with 3 startups to ~300K users each and have helped small businesses move their infrastructure online.
to register : meetu.ps/e/QkCZR/19NFR3/i
I went on a date with a girl I met on Hinge. The date went really well we talked for hours, she told me she liked me, and even said she loved the time we spent together. So naturally, I assumed there would be a second date. But the problem was (1/5) ..........
That’s one of the core things a message queue provides: the producer can keep sending 10K requests/sec, while the consumer processes 2K/sec. The remaining work becomes backlog instead of overwhelming the consumer. (4/5)