Open to AI/ML & Back-End Engineering roles

Passionate about creating value through technology.

A Computer Science graduate specializing in AI/ML & Back-End Engineering

View my workGet in touch
How I live
BUILDDEPLOYReactFastAPINode.jsKafkaTemporalML inferencePostgreSQLGitDockerKubernetesAWS
01About

Engineer at the model–system boundary

I design and build scalable server-side systems, integrate AI/ML models into production applications, and ship full-stack solutions end-to-end. Right now I'm working on an enterprise platform for orchestrating autonomous AI agents.

I care about the unglamorous parts that make AI useful in production: clean APIs, durable workflows, observability, and systems that stay reliable under load. My favorite work sits exactly where machine learning meets real backend engineering.

Lahore, Pakistandev.huzaifa.rauf@gmail.com

Back-End

  • Python
  • Django
  • FastAPI
  • Flask
  • Node.js
  • Express.js

Front-End

  • JavaScript
  • React.js
  • HTML
  • CSS

Databases

  • PostgreSQL
  • MySQL
  • MongoDB
  • ClickHouse

ML & Infra

  • Machine Learning
  • Docker
  • Kubernetes
  • AWS EC2
  • Temporal
  • Kafka
  • Git
02Experience

My career, made interactive

Instead of telling you what I've worked on, I tried to let you experience it. Each role has a small interactive demo. Try it!

Devsinc

Associate Software Engineer

Jan 2026 — Present

Lahore, Pakistan · On-site

Wand AI — Enterprise AI Workforce Platform

Contributing to an enterprise platform that orchestrates and manages the full lifecycle of autonomous AI agents inside hybrid human–AI workflows. I work across backend services, distributed workflows, and APIs that power scalable automation, intelligent task execution, and production reliability.

  • Build and maintain backend services & APIs for agent orchestration.
  • Work with durable, distributed workflows for long-running agent tasks.
  • Contribute to self-improving agents that learn from conversation history and satisfaction signals.

Self-improving agent

Support Agent

scripted demo

Agent v1 · base

Pick a question below to talk to the agent.

Self-improvement loop

Current tools

Knowledge Base

The base agent is limited. Ask a question, then run a self-improvement cycle — it learns from conversations + CSAT and upgrades itself.

Innovarix Systems

Software Engineer

Aug 2024 — Oct 2025

Remote

Web & Mobile Product Development

Worked at a fast-paced startup delivering web and mobile products for clients — building marketing websites and digital brand presences, and automating sales workflows to boost online conversions. Collaborated across the frontend and backend to ship MVPs and prototypes with full project ownership.

  • Built and shipped client websites and digital brand presences.
  • Automated sales workflows to grow online conversions.
  • Owned features end-to-end across web and mobile.

Web & mobile apps

proboxpackages.com

Overview

Orders

812+11%

Revenue

$41k+9%

Quotes

156+18%

Revenue trend

30d

Pro-Box Packages

A packaging brand's full web presence — marketing site, CMS dashboard, and a mini storefront.

AALNO AI
Hi! I'm ALNO AI — ask me anything.

ALNO AI

All-in-one AI app: custom chatbots, image generation, and PDF chat — powered by GPT-4o & Claude.

CodSoft

Machine Learning Intern

Apr 2024

Remote

ML Fundamentals on Real Datasets

Explored the Python ML stack — Pandas, NumPy, Matplotlib, and scikit-learn — working on real-world datasets in Jupyter notebooks and building a solid, hands-on understanding of machine learning fundamentals.

  • Cleaned and analyzed real-world datasets with Pandas & NumPy.
  • Trained and evaluated models with scikit-learn.
  • Visualized results and model behavior with Matplotlib.

Train a model

click anywhere to add a data point

6

points

0

epoch

4.16

loss

A linear model learns a line that minimizes the squared distance (the red residuals) to your data. Hitting Fit runs that optimization — watch the loss fall as the line settles into place.

Villaex Technologies

Python Developer Intern

Oct 2023 — Dec 2023

Lahore, Pakistan · On-site

Backend Web Development with FastAPI

Gained hands-on experience building backend web services with Python and FastAPI, while developing proficiency with Git/GitHub and a working understanding of the software development lifecycle (SDLC).

  • Built REST endpoints and services with FastAPI.
  • Practiced clean Git/GitHub collaboration workflows.
  • Learned the SDLC inside a production-minded team.

API playground

Endpoints
GET/api/health

Response

Hit Execute to send the request.

03Education & Research

Neural compression, explained simply

Traditional codecs compress every image using the same fixed set of mathematical rules. Neural compression takes a different approach by learning which information is most important to preserve rather then applying fixed compression, enabling higher visual quality at lower bitrates.

Bachelor of Computer Science

GIFT University

CGPA 3.72 / 4.0
Neural compressedNeural180 KB
JPEG compressed
JPEG180 KB

Drag to compare. At an identical 180 KB, the neural codec keeps detail that JPEG smears into blocks.

Final Year Project

Neural Image & Video Compression

My final-year project: a deep-learning compression system that outperformed traditional codecs in efficiency while preserving high visual quality, shipped as a full-stack web app

FastAPIReact.jsFirebasePyTorch

Learns what matters

Trained on real images to keep the details your eye actually notices — and discard the rest.

More quality per byte

Smaller files at the same perceived quality compared to JPEG or H.264.

Content-adaptive

Adapts to each image instead of applying the same fixed math everywhere.

Future-proof

Gets better as the models get better — no brand-new file format required.

How it works

Instead of storing every pixel, the network learns a compact representation of the image. During optimization, it gradually adjusts its parameters until it can accurately recreate that specific image using as little information as possible. The learned representation is then used to reconstruct the image whenever it is decoded.

inputencodelatentdecodeoutput

Controlling quality vs. size

The final result depends on several factors—including the size of the latent representation, the network architecture, and the balance between compression and reconstruction quality. Performance is commonly measured using PSNR, which indicates how closely the reconstructed image matches the original.The clever part: at any file size, this network keeps more quality than older formats like JPEG, because it learned which details actually matter to your eyes.

smaller file · lower PSNRlarger file · higher PSNR
Contact

Let's build something reliable.

Open to AI/ML & Back-End Engineering roles. Whether it's AI agents, backend systems, or a full-stack build — I'd love to hear what you're working on.

Email me