Full Stack Engineer · AI integrations

I ship AI-powered products
that hold up in production.

React, TypeScript, Node, NestJS. LLM integrations that work, measured, and open source.

production.ts
const app = createApp();
app.use(aiReview(), {
  queue: "bullmq",
  model: "kimi-k2",
});
// 92% accuracy over 500+ resumes
// 8.2x throughput in production
export default app;
8.2×throughput, Hono vs Express
53×lower latency, published
92%ATS scoring accuracy, 500+ resumes
5+merged PRs, open source
Usama Mangi
Based in Lahore · Open to remote

About

Build it, ship it, document it.

I'm Usama Mangi, a full stack engineer in Lahore, Pakistan. For about 18 months I've shipped AI-powered products. An AI resume analyzer scored 500+ resumes at 92% accuracy. A code-review bot reviews pull requests with an LLM, in a queued production pipeline. A marketplace app, Matasha, is live on the App Store and Play Store.

I publish what I learn. My Hono-vs-Express benchmark documented 8.2x throughput and 53x lower latency. When the AI hype cools, what's left is code that works. I build that.

Experience

Employers, dates, shipped work.

Full-Stack AI Engineer

ZynSols

Aug 2025 –PresentSukkur, Pakistan · 1.1 years

Shipped an AI resume analyzer that scored 500+ resumes at 92% accuracy, a marketplace app with Stripe processing, and the published Hono-vs-Express benchmark (8.2x throughput).

React Native App Developer Intern

Indus Tech Nexus

Jun 2026 – Jul 2026Lahore, Pakistan · 2 months

Integrated Stripe payments, built two Express + PostgreSQL REST APIs handling 10,000 requests/day, and refactored 10+ legacy class components.

Open-Source Contributor

GirlScript Summer of Code

Oct 2024 – Nov 2024Remote · 2 months

Merged 5+ pull requests across 3 React and Node.js projects, shipping end-to-end MERN features through structured code review.

B.S. Computer Science

Sukkur IBA University

Sep 2022 – May 2026Sukkur, Pakistan · 3.8 years

Core CS, with a final-year focus on self-supervised segmentation (SimCLR / MoCo) and generative image restoration.

Full history and dates are on the downloadable resume.

What I do

Four ways I add to a team.

01

Full stack product engineering

React, TypeScript, Node, NestJS. End to end features with auth, clean APIs, and queues that don't drop events.

02

AI + LLM integration

Document parsing, RAG, agents, code review. LLM hooks that hold up in production, measured at 92% accuracy across 500+ resumes.

03

Performance engineering

Profiling, load testing, latency budgets. I publish the numbers I measure, and the 8.2x RPS gap is in a public benchmark you can rerun.

04

Mobile with React Native

Expo, payments, push notifications, geolocation. Matasha is live in both app stores.

Skills

The stack I work on.

Languages

  • TypeScript
  • JavaScript
  • Python
  • SQL

Frontend

  • React
  • Next.js
  • React Native
  • Expo
  • Tailwind CSS
  • Framer Motion

Backend

  • Node.js
  • NestJS
  • Express
  • Hono
  • Bun
  • Flask

Data & AI

  • PostgreSQL
  • MongoDB
  • Redis
  • OpenAI API
  • LLM integration
  • PyTorch
  • scikit-learn
  • Prompt engineering

Infrastructure

  • Docker
  • GitHub Actions
  • Nginx
  • BullMQ
  • Postman

AI tooling

  • Claude Code
  • Cursor
  • GitHub Copilot

Selected projects

Shipped and documented.

01open source

Career Autopilot

Full Stack · AI/ML · HR Tech · SaaS

AI-powered job portal: NestJS + React 19 + OpenAI + BullMQ + PostgreSQL. Resume analyzer scores 500+ resumes at 92% accuracy.

4 technologies · 2026

Open the repo
Career Autopilot
Matasha

02live

Matasha

Mobile · Marketplace · Real-Time · Stripe

Geo-based flooring marketplace for UK homeowners. React Native + Expo + Firebase + Stripe. Live in both stores.

See it live
AI Code Review Bot

03live

AI Code Review Bot

AI/ML · DevTools · GitHub App · Async Pipeline

GitHub App that reviews PRs with an LLM. Express + BullMQ + Redis + PostgreSQL + Kimi-K2. Async review pipeline, inline diff comments.

See it live
Hono vs Express benchmark

04open source

Hono vs Express benchmark

Performance · Backend · Benchmarking · Bun

Identical MVC REST API on Hono and Express 5, both on Bun. Autocannon: 8.2x RPS, 53x lower latency. Open source.

Open the repo

Certificates

Verifiable certifications from Google, Meta, and Anthropic

LinkedIn

Creating and Hosting a Full-Stack Site

Issued Jun 2025

LinkedIn

MERN Essential Training

Issued Nov 2024

Case Studies

Posts and notes.

Self-supervised segmentation, after the paper hype

SimCLR and MoCo pretraining gave me a ResNet50 that worked. Here is what I would do differently if I were starting over today.

For my final-year project I pretrained a ResNet50 with SimCLR and MoCo on a corpus of medical images, then fine-tuned it for segmentation against a fully supervised baseline. The self-supervised encoder won on the small labelled set, lost on the large one, and the crossover point was somewhere around 2,000 labelled images.
Read the post

Scoring 500 resumes at 92% accuracy — what I learned about prompts

What the production run of an LLM-based resume analyzer taught me about prompt stability, edge cases, and the cost of bad input.

The resume analyzer hit 92% accuracy on a hand-labelled set of 500+ resumes. That number came after a long stretch of lower numbers, and the gap between the two states was almost entirely about how the prompt handled bad input.
Read the post

Why I benchmarked Hono against Express on Bun

How the same MVC REST API measured 8.2x RPS and 53x lower latency, and what that does and does not tell you about your own stack.

I built the same MVC REST API twice: once on Hono, once on Express. Same routes, same middleware, same JSON payloads, same database layer. The only variable was the framework. Then I ran autocannon at 200 concurrent connections for 30 seconds against both, on the same Bun runtime, on the same machine.
Read the post
Read all posts

Contact

Open to software engineering roles.

On-site in Karachi or Lahore, or remote. Willing to overlap US hours; EU hours come naturally. I reply within 24 hours.