Available for work

Abdullah
Bakr

// AI/ML & Deep Learning Engineer

Building real intelligent systems — from convolutional networks to published open-source libraries. 3+ years in Python, 1+ year shipping AI projects that people actually use.

View Projects LinkedIn

Scroll to explore

deepcsv.py ● live

1# pip install deepcsv

3from deepcsv import process_file, read_any, clean_values, auto_fix

5# ─── process_file() ────────────────────────────

6# Converts string lists → NumPy arrays, fixes mixed types

7process_file("data.csv")

8process_file("data.csv", file_format="parquet")

10# ─── process_all_files() ───────────────────────

11# Batch processes entire folder trees recursively

12process_all_files("path/to/folder", file_format="csv")

14# ─── read_any() ────────────────────────────────

15# One universal reader — csv, xlsx, json, parquet…

16df = read_any("data/users.csv")

17df = read_any("warehouse/orders.parquet")

19# ─── clean_values() ────────────────────────────

20# Cleans nulls, values, types with full control

21clean_values(df, cols=["age", "salary"])

22clean_values(df, finding_type=str, cols=["score"])

24# ─── auto_fix() ────────────────────────────────

25# Detects & fixes mixed-type columns automatically

26df = auto_fix("data.csv")

27df = auto_fix(my_dataframe)

29# Arrays restored. Types fixed. Done.

↕ scroll to explore

python ≥ 3.7

PyPIPublished

MITLicense

OpenSource

About Me

Building Systems
That Actually Work.

I'm Abdullah Bakr, an AI/ML & Deep Learning Engineer from Egypt. I build real, working intelligent systems, not just notebooks gathering dust on Kaggle.

I've trained convolutional networks reaching ~98% accuracy on image classification, published an open-source data cleaning library DeepCSV on PyPI, and built end-to-end ML regression pipelines — all documented publicly on GitHub and Kaggle.

My foundation is strong: 3+ years with Python, solid mathematical background in linear algebra, calculus, and statistics, and hands-on experience with the full AI stack from data wrangling to model deployment.

Currently open to freelance projects, internships, and full-time roles in AI/ML.

"I write code, train models, and ship things people can actually use."

Core Skills

🐍

Python

Primary language — scripting, ML pipelines, open-source libs

🧠

Deep Learning

Neural networks, CNNs, training loops, regularization

👁️

Computer Vision

Image classification, data augmentation, OpenCV

⚡

TensorFlow / Keras

Model building, custom layers, callbacks, evaluation

🔥

PyTorch

Tensor ops, custom training loops, experimentation

📊

ML / Data Science

Regression, classification, feature engineering, EDA

Projects

What I've Built

Real-world projects across computer vision, machine learning, open-source tooling, and more — browse by category or filter by tech stack.

Showing — projects

Computer Vision ~98% Acc

01 / 12

Computer Vision · Deep Learning

Cats vs Dogs Image Classifier

Custom CNN achieving ~98% accuracy on binary image classification. Built with TensorFlow/Keras — data augmentation, dropout regularization, batch normalization, and hyperparameter tuning.

~98% Accuracy TensorFlowKerasCNN

GitHub Kaggle

Open Source PyPI ✓

02 / 12

Open Source · PyPI Package

DeepCSV — Automatic Data Cleaner

Python library published on PyPI. Auto-walks directories, converts stringified lists to NumPy arrays, fixes mixed-type columns, and saves as Parquet — built for real ML data prep.

pip install deepcsv PandasNumPyPyArrow

GitHub Kaggle PyPI

Regression Model ML Pipeline

03 / 12

Regression · Machine Learning

Weather Temperature Predictor

End-to-end regression pipeline predicting weather temperatures from historical meteorological data. Feature engineering, model comparison (Linear, Ridge, RF), cross-validation, full evaluation.

Regression Pipeline Scikit-learnPandasMatplotlib

GitHub Kaggle

Transfer Learning ~92.5% Acc

04 / 12

Computer Vision · Transfer Learning

Intel Image Classification

Multi-class scene classification using Transfer Learning with InceptionV3 on the Intel Image dataset (6 categories: Buildings, Forest, Glacier, Mountain, Sea, Street). Achieved ~92.5% validation accuracy with a frozen pretrained base and custom classification head.

~24K Images TensorFlowInceptionV3Keras

GitHub Kaggle

Transfer Learning ~93% Acc

05 / 12

Computer Vision · Classification

Garbage Classification (Inception V3)

High-accuracy image classification model for waste sorting using Inception V3 architecture. Classifies garbage into categories (cardboard, glass, metal, paper, plastic, trash) to support automated recycling systems.

~93% Accuracy PyTorchInceptionV3~25K Images

GitHub Kaggle

Medical Imaging 92% Acc

06 / 12

Medical AI · Deep Learning

Breast Cancer Classification (VGG19-BN)

Deep learning model for breast cancer classification using VGG19 with batch normalization on ultrasound images (benign / malignant / normal). Achieved 92% test accuracy with a confusion matrix showing strong generalization.

92% Test Accuracy PyTorchVGG19-BNMedical Imaging

GitHub Kaggle

Breast Cancer Classification Transfer Learning

Medical Imaging 98.5% Val Acc

07 / 12

Medical AI · Transfer Learning

Breast Cancer Classification (Transfer Learning)

Comparative study of transfer learning models (ResNet50, VGG16, EfficientNetB4) for breast cancer classification on ultrasound images. Best model (ResNet50) achieved 98.5% validation accuracy with 2-class prediction (benign / malignant).

98.5% Val Accuracy TensorFlowResNet50Medical Imaging

GitHub Kaggle

📄 Research Paper

Medical Imaging · Research 112K+ Images

08 / 12

Medical AI · Multi-Label · Research

NIH Chest X-Ray Disease Classification

Multi-label classification of 14 thoracic diseases from 112K+ chest X-ray images (NIH ChestX-Ray14). Benchmarking DenseNet-121, ResNet-50, EfficientNet-B4, and ViT-Base against a proposed Swin Transformer, with Grad-CAM explainability. Targeting peer-reviewed publication.

⚗ In Research PyTorchSwin TransformerGrad-CAM

DenseNet-121 ResNet-50 EfficientNet-B4 ViT-Base Swin Transformer ✦

GitHub — Soon Paper — Upcoming

Computer Vision · Forensics 95%+ Acc

09 / 12

Computer Vision · AI Forensics · ICANN 2026

Synthetic Image Attribution

Dual-stream EfficientNet-B4 pipeline that fingerprints AI-generated faces across 10 text-to-image models. RGB stream + SRM forensic noise filters + TTA — built for the ICANN 2026 DLMMDD Workshop Challenge.

95%+ Leaderboard PyTorchEfficientNet-B4SRMTTA

GitHub Kaggle

ML / Classification ~99.8% Acc

10 / 12

ML · Classification · Feature Engineering

Iris Flower Classification Showdown

End-to-end classifier benchmark on the Iris dataset — feature engineering from 4→14 features, SelectKBest selection, and head-to-head comparison of Logistic Regression, Ridge, Random Forest, and Gradient Boosting.

~99.8% Setosa Scikit-learnFeature Eng.SelectKBest

GitHub Kaggle

Computer Vision · Transfer Learning ~70% Acc

11 / 12

Computer Vision · Age Estimation

FaceAge Classifier

Age group classification from ~32K face images using ResNet50 & DenseNet121. Three age classes (14–24, 25–40, 41–70) with full fine-tuning, AdamW, and early stopping. DenseNet121 edges out at ~70% accuracy.

~70% DenseNet121 PyTorchResNet50DenseNet121

GitHub Kaggle

Computer Vision · Transfer Learning AUC 0.9997

12 / 12

Computer Vision · Safety · Binary Classification

Drowsiness Detection — Eye State Classifier

Binary classification of infrared eye images (Awake vs Sleepy) on the MRL Eye Dataset (~85K images). ResNet50 hits AUC 0.9997, InceptionV3 hits 0.9993 — both trained with mixed precision and early stopping.

AUC 0.9997 PyTorchResNet50InceptionV385K imgs

GitHub Kaggle

🔍

No projects found

Try selecting a different category or clearing the filter.