Portfolio
Featured Projects
AI/ML and full-stack projects built during my Master's research at UM-Dearborn.
Deepfake Detection with CNN Attention
Comparative study of CNN architectures enhanced with SE Attention and CBAM for binary classification of authentic vs. AI-generated facial images. Achieved 91.7% validation accuracy.
Market Basket Analysis with PySpark
Scalable association rule mining on 3.2M+ retail transactions using Apache Spark FP-Growth. Extracted 436 rules, max 73.8× lift, enabling data-driven inventory recommendations.
Charge-Aware EV Trip Planner
Battery-constrained shortest path using Dijkstra's algorithm and a dynamic programming battery state table across a 10-city road network. Runs in 0.05ms.
ConvNeXt vs EfficientNet Deepfake Detection
Two-stage transfer learning on 190,335 facial images. ConvNeXt-Tiny achieved 92.9% test accuracy with Grad-CAM explainability, beating EfficientNetV2-S by 6.35 pp.
FinTrack Financial Planner
Full-stack expense, budget, and savings-goal tracker. React 19 + Express/Prisma REST API + PostgreSQL + JWT auth. Two-tier budget enforcement with Swagger docs.
Expertise
Skills & Technologies
Machine Learning & AI
Data Science & Engineering
Tools & Platforms
Background
Education & Research
Academic background and AI/ML research projects conducted at UM-Dearborn.
Education
M.S. in Artificial Intelligence
University of Michigan–Dearborn, Dearborn, MI, USA
Relevant Coursework: Deep Learning, Big Data (Hadoop & Spark), Python for Data/ML, Algorithms.
B.S. in Computer Science
Pir Mehr Ali Shah Arid Agriculture University (PMAS-AAUR), Rawalpindi, Pakistan
Bachelor's degree in Computer Science covering core CS fundamentals, software engineering, databases, and programming.
Research & Projects
ConvNeXt vs EfficientNet for Deepfake Detection
CIS 579 · AI Research · UM-Dearborn
Two-stage transfer learning study on 190,335 facial images. ConvNeXt-Tiny achieved 92.9% test accuracy. Applied Grad-CAM explainability to analyze spatial attention patterns in correct and misclassified samples.
Deepfake Detection with CNN Attention
CIS Research · UM-Dearborn
Designed and implemented a deepfake detection system using CNN with SE Attention and CBAM. SE Attention raised accuracy from 82.4% to 91.7%. Analyzed precision/recall trade-offs with data augmentation.
Market Basket Analysis using PySpark
CIS Data Engineering · UM-Dearborn
Built a scalable association rule mining pipeline on 3.2M+ retail transactions using Apache Spark FP-Growth. Extracted 436 high-confidence rules with a maximum lift of 73.8×.
Charge-Aware EV Trip Planner
CIS 505 · Algorithms · UM-Dearborn
Battery-constrained shortest path planner using Dijkstra's algorithm and a dynamic programming battery state table across a 10-city road network. Executes in 0.05ms.
Contact
Let's Work Together
Open to ML internships, junior roles, and research collaborations. Drop me a message and I'll reply fast.