h
hananali5

Hanan Ali Khan

@hananali5

AI Engineer, Computer Vision Specialist, Python Developer

Pakistan
English, Urdu
About me
I am an AI Engineer specializing in Machine Learning, Deep Learning, Computer Vision, and Python development. I build AI solutions including deepfake detection, image and video analysis, object detection, automation tools, and custom machine learning models. Skilled in Python, PyTorch, TensorFlow, and OpenCV, I help businesses and researchers turn ideas into reliable, scalable, and production-ready solutions. I focus on quality, clear communication, and timely delivery.... Read more

Skills

h
hananali5
Hanan Ali Khan
Offline • 

See my services

AI Websites & Software
I will build ai, machine learning and computer vision solutions
Scripting
I will automate tasks using python scripts, bots and workflows

Portfolio

Work experience

Fiverr

Final Year Project — Machine Learning Researcher

Fiverr • Full-time

Jun 2025 - Present1 yr 1 mo

Project: “Dual-Domain Synthetic Media Detection: From Gated Image Fusion to Spatial-Temporal Video Pipelines” Executed a 6-stage deep learning framework for detecting deepfakes, GAN-generated images, and facial manipulations across images and videos. 📸 Image-Based Deepfake Detection (Spatial + Frequency Domain) • Developed a dual-domain framework combining RGB spatial features with 2D Fast Fourier Transform (FFT) frequency analysis to detect hidden synthetic artifacts. • Implemented a learned Gated Fusion architecture using parallel EfficientNet-B0 streams with adaptive sigmoid-based feature weighting. • Achieved 94.40% Accuracy and 0.9468 F1-Score on a balanced dataset of 30,000 face images. 🎥 Video-Based Spatial-Temporal Detection Pipeline • Built an end-to-end deepfake video detection pipeline optimized for compressed real-world media. • Combined an ImageNet-pretrained Xception backbone with a 2-layer BiLSTM for temporal sequence modeling. • Integrated a Temporal Attention mechanism for dynamic frame weighting and sequence-level representation learning. • Engineered a GPU-accelerated preprocessing pipeline using MTCNN face extraction and standardized 16-frame video sequencing. 🛠️ Optimization & Failure Analysis • Diagnosed optimization collapse and shortcut-learning behavior in RGB+FFT temporal fusion models, successfully pivoting to a more stable RGB spatial-temporal architecture. • Addressed frame imbalance and compression inconsistencies using weighted BCEWithLogitsLoss optimization. • Applied Temperature Scaling during inference to improve probability calibration and reduce overconfident predictions. 📈 Final Video Benchmark (FaceForensics++ c23) • Test Accuracy: 82.47% • ROC-AUC: ~0.93 • Generalized across DeepFakes, FaceSwap, Face2Face, and NeuralTextures manipulations. Tech Stack: PyTorch, TensorFlow, OpenCV, MTCNN, 2D-FFT, Xception, EfficientNet-B0, ResNet50, BiLSTM, Temporal Attention, AMP, AdamW, Django, Flask, Python.

Upwork

Shopify Store Developer

Upwork • Freelance

Dec 2024 - Jan 20251 mo

Designed and developed a fully functional e-commerce platform for a luxury fashion brand. Built store architecture from scratch, customized Shopify themes, and implemented UI/UX improvements. Structured product catalogs and optimized inventory management while improving website loading speed and mobile responsiveness.

Fiverr

Lead Generation Executive

Fiverr • Freelance

Jan 2024 - Apr 20243 mos

Generated qualified leads through targeted outreach and data-driven prospecting strategies. Analyzed campaign performance and customer engagement data to optimize lead generation processes. Designed outreach workflows and utilized analytical tools to measure performance and refine strategies for continuous improvement.