Repository → containersConcept
Halvo
Connect a GitHub repository and Halvo detects or generates a Dockerfile, builds the image, and streams the build log over a WebSocket while the container comes up.
Good systems.Built from theground up
I’m Ankit Kumar, also known as deblomper, a full-stack developer who likes to get under the hood. I build deployment platforms, backend systems, and things that keep running.
Open to good opportunities
Independent thinking. Hands-on engineering.Building at nub-coders since 2020
Selected work.
Ideas turned into tools.
Built, shipped, and kept running.
Your domain, your emailConcept
NubMail
A self-hosted mail stack — Postfix out, Dovecot for IMAP and POP3, a hand-written SMTP receiver in — with a dashboard for domains, mailboxes and API sending.
Request → mediaConcept
Ytube API
An async FastAPI service for YouTube search, metadata and stream resolution that runs an embedded Deno runtime to evaluate signature challenges, with Redis token buckets in front.
Queue → group voice chatConcept
Nub Music Bot
Streams HD audio and video into Telegram group voice chats from a pool of assistant clients, with address-level SSRF checks on every remote URL and a sweeper for idle calls.
Built with
a point of view.
Some things matter long after the first deployment.
Operate what you build
I maintain the services I build — Halvo, NubMail, and Ytube API run live 24/7 on my own infrastructure. Real operational feedback (container leaks, SMTP quirks, upstream API breaking changes) drives reliable design far better than local mocks.
Strict input validation at the boundary
Validate against real attack vectors: resolve IP addresses to block SSRF before fetching remote URLs, enforce byte-and-entry quotas during live archive extraction, and verify DNS records (SPF, DKIM, DMARC) before sending mail.
Automate rules where code runs
If an error is easy to introduce and silent at runtime — like un-awaited async database queries in MongoDB — enforce it via static AST checks in CI. Don't rely on human review to catch silent failure modes.
Understand the layer beneath
Building a key-value store in Zig or TDLib bindings in C++20 isn't about reinventing wheels — it's about knowing exactly how memory allocators, hash tables, and event loops behave when high-level abstractions fail.
A little more
behind the code.
Building in public since 2020.
I am a full-stack developer working mostly on backend systems and the infrastructure underneath them. TypeScript and Python are where most of the work lives; Zig and C++20 are where I went when I wanted to understand a layer instead of consume it.
The through-line is operational. Almost everything here is a long-running process talking to something it does not control — the Docker daemon, an SMTP peer, Telegram's calling layer, an upstream that rewrites its JavaScript without notice. Most of what I know came from those seams rather than the application code on either side.
Published under nub-coders since 2020. This site shows nine of twenty-six repositories; forks, experiments and early code are left out on purpose.
Explore nub-codersMy working toolkit
- Languages
- TypeScript · Python · Zig · C++20 · Go · Rust · C
- Runtime & infra
- Docker · Compose · Traefik · Kubernetes · Postfix · Dovecot · nginx
- Data
- PostgreSQL · MongoDB · Redis · Prisma
- Interfaces
- React · Next.js · Tailwind · FastAPI · Express · Pyrogram · TDLib
Have something
in mind?
Let’s talk systems, infrastructure, or what you’re building next. Available for full-time backend roles, contract systems engineering, and collaborations on operations tooling.
Open to new opportunities
Drop me a note.
A question, an idea, or a simple hello.