~/priyanshu-kumar-singh · backend & systems

PRIYANSHU KUMAR SINGH

Backend & systems engineer-in-training. I build things that stay fast on small machines — containers, kernels, queues, and everything between the syscall and the status code.

Dehradun, IN · BCA ’27 @ DIT · open to internships & new-grad SDE roles

0rps sustained

0p95 latency

0requests served

0leetcode solved

→ inspect projects $ ./contact.sh

priyanshu@dehradun: ~/portfolio

$ cat about.md

Systems-minded, backend-first

──────────────────────────────────

I’m a backend & systems-focused developer doing my BCA at Dehradun Institute of Technology (expected May 2027). I like the parts of software most people treat as plumbing: process lifecycles, file descriptors, memory limits, TLS handshakes — and what actually happens when you run five services on a 1 GB VM.

Most of what I build is shaped by constraints. I squeezed a FastAPI + Redis + Celery + PostgreSQL stack onto a single tiny Azure VM, sandboxed untrusted code in Docker, and kept it responsive at 81+ RPS with 11ms p95 across 51K+ requests.

When I go lower level, I write shells in C — fork(), execvp(), dup2() — track every heap allocation by hand, and read the man page before the Stack Overflow answer.

neofetch

priyanshu@dehradun

hostDehradun Institute of Technology
degreeBCA · expected May 2027
baseDehradun, Uttarakhand, IN
stackPython · JavaScript · C
focusbackend · systems · infra
kernelcuriosity (rolling release)
shellabyss — hand-rolled, in C
uptimeshipping under constraints

$ apt list --installed

The toolchain

Everything below has been used in a real build — not just watched in a tutorial.

┌─[ languages ]─┐

PythonJavaScriptC

┌─[ backend ]─┐

FastAPINode.js

┌─[ databases ]─┐

PostgreSQLRedis

┌─[ systems & infra ]─┐

LinuxDockerDocker ComposeAzureNginx/CaddyCelery

┌─[ networking & security ]─┐

HTTP/HTTPSSSL/TLSDNSCORSJWT/JWKS

┌─[ frontend ]─┐

Next.jsReact.jsTailwindCSS

$ ls ./projects --sort=impact

Selected builds

Two projects, read top to bottom: one runs a full production-ish stack on a 1 GB VM, the other lives one syscall above the kernel.

Containerized Code Execution Platform

FastAPI · Next.js · Docker · Redis · Celery · PostgreSQL

↗ live ⌥ github

0rps sustained

0p95 latency

0requests

0concurrent sandboxes

1 GBtotal vm ram · zero oom kills

// architecture

// load test · p95 ≈ 11ms across run

A full stack on 1 GB of RAM

Deployed the containerised execution backend on an Azure Ubuntu VM with 1 GB RAM; configured custom swap space and Docker Compose so FastAPI, Redis, Celery and PostgreSQL coexist under memory pressure — the box handles 50+ concurrent sandboxed submissions without a single OOM kill.

HTTPS-only, end to end

Put Caddy in front as the reverse proxy with automated Let’s Encrypt SSL/TLS and locked Azure NSGs to HTTPS-only routing. Tracked down and resolved the CORS and Mixed Content breakage between the Vercel-hosted frontend and the Azure backend.

Isolation by containers

Designed the execution model around per-container CPU/memory limits and Docker-based process isolation, so hostile or buggy submissions can’t touch the host — or each other.

Off the event loop

Used Redis + Celery to offload execution tasks from the FastAPI event loop; submissions become queued jobs instead of blocking coroutines, keeping the API responsive under load.

Zero-trust entry

Validated JWTs against an external JWKS endpoint before processing any execution request — unauthenticated traffic is blocked at the door.

State you can watch

Built a long-polling Next.js client that streams container state transitions into a React terminal UI, with adaptive polling intervals to keep server load down. QUEUED→RUNNING→SUCCESS/ERROR

The receipts

Sustained 81+ RPS with 11ms p95 latency across 51K+ requests on a constrained 1 GB Azure VM — by decoupling heavy execution work through Redis and Celery.

Abyss Shell

C · Linux Kernel API · POSIX

⌥ github

0memory leaks

0segfaults

0repl test iterations

2phase tokenizer

// pipeline internals

// live repl demo

./abysh

A shell, from syscalls up

Built a Unix shell in C using fork(), execvp() and waitpid() for process creation, execution and lifecycle management across a persistent REPL.

Pipelines, not just commands

Implemented multi-pipe command execution for chained pipelines, dynamically managing file descriptors and IPC across child processes.

Redirection: > and >>

Added I/O redirection via dup2() and open(), supporting both > (overwrite) and >> (append) modes.

A tokenizer that doesn’t lie

Wrote a two-phase tokenizer to handle nested pipe contexts where strtok caused state collisions — parsing that survives real input.

Valgrind-clean

Zero memory leaks and zero segfaults across 500+ REPL test iterations; every heap allocation tracked and freed manually.

$ cat education.log

Education

──────────────────────────────────

expected May 2027

Bachelor of Computer Applications (BCA)

Dehradun Institute of Technology — Dehradun, India

// relevant coursework

Data Structures & Algorithms System Programming OS Internals Computer Networks

$ leetcode --stats

Problem solving

Pattern recognition is a muscle. These are the reps.

problems solved · in python

✓ NeetCode 150 — full set completed

Including the complete NeetCode 150 — arrays, trees, graphs, dynamic programming and sliding window.

// coverage

[✓]Arrays & Hashing
[✓]Trees
[✓]Graphs
[✓]Dynamic Programming
[✓]Sliding Window

$ ./reach_out.sh

Let’s build something

Hiring a fresher who reads man pages for fun and ships under constraints? Ping me.

mailto://pksingh.backend@gmail.com

tel://+91-6203617455

git://github.com/yansh07

ln://linkedin.com/in/yansh08

Dehradun, Uttarakhand · open to internships & new-grad backend/SDE roles · relocatable & remote-ready.

$ mail -s "opportunity" pksingh.backend@gmail.com