About
Who I am
I build AI systems that leave the notebook. My work ends with something running: a barrier arm that responds to a verified identity, a detector optimized for edge hardware, or a screening workflow an HR team can use in its daily process. The distance between a model that works and a system that works is where I spend my time.
I'm an AI engineer with over four years in the industry, currently at Impact Business Solutions in Islamabad. My M.S. Artificial Intelligence work at NUST CEME includes deep learning, NLP, computer vision, embedded AI, and cybersecurity; I ranked first in the program with a CGPA of 3.92 out of 4.00.
I've delivered work for national and international clients across computer vision, edge deployment, document intelligence, and LLM applications. It runs from research — a two-stage intrusion detection system benchmarked across 2.8 million network flows, written up as an IEEE-format paper — to systems I can point at, like a CNIC-reading access control unit that has been deciding who gets into a parking facility since the day it went live.
- CURRENT
- AI Engineer — Impact Business Solutions
- EXPERIENCE
- 4+ years in industry
- EDUCATION
- M.S. Artificial Intelligence — NUST CEME
- STANDING
- Ranked 1st in program · CGPA 3.92 / 4.00
- COURSEWORK
- Deep learning, computer vision, NLP, embedded AI, cybersecurity
- FOCUS
- Computer vision · edge AI · document intelligence · LLM systems
- BASED
- Islamabad, Pakistan (UTC+5)
“A model that scores well in a notebook and a system that works at a barrier in the rain are two different achievements.”
Achievements
- Ranked first in the MS Artificial Intelligence program at NUST CEME, CGPA 3.92/4.00.
- Systems in production, not prototypes — including an access-control unit processing live vehicle entry and an HR screening tool in daily use.
- Published research — an IEEE-format paper on cascaded network intrusion detection, co-authored with Dr. Muhammad Shafiq (DC&SE, CEME, NUST).
- Real-time inference on constrained hardware — YOLO detection at 100ms or under per frame on a Jetson Nano, via TensorRT and reduced-precision quantization.
How I think about the work
Deployment is the deliverable.
A model that only runs on my machine isn't finished. The work is done when it runs on yours, on your hardware, against your data.
Report the number that isn't flattering.
Every optimization costs something. I measure what quantization did to accuracy and tell you the figure, because a client who only ever hears good news has no way to tell whether to believe me.
Constraints are the interesting part.
Making a detector fast enough for a hundred-dollar device is harder, and worth more, than adding a percentage point of accuracy on a GPU that costs more than the vehicle it's watching.
The system has to survive me.
Documentation and a handover your team can actually work from. If a project leaves a client dependent on me to keep it alive, I've built it wrong.
Experience
Jun 2026 – Present · Islamabad, Pakistan
AI Engineer
Impact Business Solutions
Contributing to AI-powered products, automation workflows, and intelligent business systems.
- Apply machine learning, LLMs, and data-driven technologies to practical business problems.
- Support digital-transformation initiatives across AI-powered business systems.
Aug 2025 – Jan 2026 · Islamabad, Pakistan
AI Engineer (Contract)
RISETech
Delivered an end-to-end machine-learning solution for a corporate client.
- Owned the solution from data-pipeline optimization through enterprise deployment.
Jan 2024 – Jun 2026 · Remote
Freelance AI & Machine Learning Engineer
Independent
Designed and deployed AI solutions for independent clients.
- Built LLM applications, RAG pipelines, and computer-vision models for client projects.
- Delivered chatbots, automation systems, data analysis, and predictive-modeling workflows.
If you're weighing up a system and want to know whether it's feasible before committing a budget, I'm happy to tell you — including when the answer is no.