Underwater trash detection
I fine-tuned a YOLOv8s segmentation model on the TrashCan underwater dataset and served it through a Flask app with filename sanitising and a file-type allowlist. It reached 0.712 box mAP@50 after a 10-epoch baseline run.
I'm a computer science graduate who builds network security tools and then tests where they fail. I'm looking for an entry-level security role and working toward offensive security and AI red teaming.
A real-time pipeline that captures live traffic, scores each flow with machine-learning models trained on CICIDS2017, and pushes enriched alerts to a dashboard. I led the engineering on a three-person team. Run either test below to see what happened.
Random Forest F1 on a held-out CICIDS2017 split of 45,149 flows. DDoS traffic only.
Replay recall at 100% precision, up from 52% after I found and fixed three mismatches in my feature extractor.
Rate rule I added after the models missed a real flood from my phone.
pytest tests covering preprocessing, train/test leakage, fusion logic, route security and blocking.
I fine-tuned a YOLOv8s segmentation model on the TrashCan underwater dataset and served it through a Flask app with filename sanitising and a file-type allowlist. It reached 0.712 box mAP@50 after a 10-epoch baseline run.
As a web development and database intern I built responsive intranet pages, designed the SQL schemas behind the department's internal reports, and wrote a JavaScript time-tracking tool staff used.
Walkthroughs of the labs I solve, with what I tried, what failed and how I got in.
Writeups of the PortSwigger labs I solve: what I tried, what failed, and what got me in. The first ones go up in November 2026.
Notes from the rooms I complete alongside the PortSwigger labs.
My plan from here to an offensive security role.
Based in Peshawar. Open to entry-level security and IT roles: on-site in Pakistan, remote, or relocating abroad. Found the flag? Email it to me.