i
i_am_sabi

Abdul Saboor

@i_am_sabi

Medical AI and Computer Vision Specialist

Paquistão
Inglês, Hindi
Algumas informações são exibidas no idioma inglês.
Sobre mim
I am a passionate AI, Machine Learning, and Deep Learning Engineer with over 3 years of hands-on experience building intelligent solutions. Specialized in Python, pandas, and data analysis, I have a proven track record of developing high-accuracy predictive models—such as heart disease prediction systems utilizing XGBoost and academic analytics platforms (EDUAI). Holding a B.Sc. in Software Engineering, I bridge the gap between complex data and actionable, automated business solutions. Let's work together to turn your data into intelligence.... Saiba mais

Habilidades

i
i_am_sabi
Abdul Saboor
offline • 

Conheça meus serviços

Machine learning
I will develop advanced deep learning and computer vision pipelines
A partir deUS$ 150 /projetoMais detalhes

Portfólio

Experiência profissional

Freelancing_Career

Deep Learning Engineer | Medical Computer Vision Freelance / Self-Employed

Freelancing Career • Freelance

Aug 2022 - Present • 4 yrs 1 mo

Engineered an end-to-end, production-grade medical computer vision pipeline to automate the classification and localization of pneumonia opacities using the RSNA Challenge dataset. Optimized natively for local GPU hardware (NVIDIA CUDA, RTX 2060 SUPER) with an exceptional single-image inference speed of 0.9ms. Key Contributions & Engineering Phases: • Medical Data Engineering: Extracted and parsed 12-bit/16-bit DICOM (.dcm) files via pydicom; merged complex metadata spreadsheets into a unified patient target matrix. Enforced data integrity using MD5 cryptographic hashing algorithms to eliminate corrupted data. • Preprocessing & Augmentation: Implemented CLAHE (Contrast Limited Adaptive Histogram Equalization) to contrast-boost faint lung fluid pockets. Managed image downscaling while mathematically maintaining exact 1:4 bounding box coordinate scaling ratios. Applied Z-Score normalization and robust clinical augmentations (torchvision.transforms.v2). • Multi-Model Architecture: Built a custom Dual-Headed Network over a pre-trained ResNet backbone, splitting output features into a binary classification branch and a 4-coordinate bounding box regressor optimized with Smooth L1 (Huber) Loss. • Next-Gen Ensembling (2026 Architectures): Re-architected and upgraded the pipeline to incorporate self-attention Swin Transformers (via timm registry) and real-time YOLO11 object detection loops. • Validation & Post-Processing: Mitigated data leakage using a strict 5-Fold Cross-Validation split grouped by unique patient ID. Engineered an evaluation framework to calculate mean Average Precision (mAP) across 8 official competition IoU thresholds (0.40 to 0.75). Fused overlapping predictions utilizing Non-Maximum Suppression (NMS) and Weighted Boxes Fusion (WBF).

Systemized_Business Automation - Achieve More & By Doing Less

Freelance AI/ML/DL Engineer | Building Intelligent Solutions

Systemized Business Automation - Achieve More & By Doing Less • Meio período

Apr 2021 - Apr 2024 • 3 yrs

* Developed a Heart Disease Prediction Model with 88% accuracy using Linear & Logistic Regression and 95% using XGBoost, improving early risk detection. * Built EDUAI, an intelligent school analytics platform using regression, boosting, and clustering models to predict student performance and progress based on multiple academic and behavioral factors.