Soliman Zakaria

Driven Computer Science graduate with strong foundations in software engineering, full-stack development, and applied AI. Experienced in developing scalable systems, optimizing algorithms, and building deep learning models using TensorFlow and PyTorch. Recognized for innovative problem-solving and measurable impact in internships. Seeking an AI or software engineering role where I can build intelligent, data-driven systems that deliver real results.

Neural Network for Image Classification (MNIST & CIFAR-10)  

Oct 2023 – Jan 2024

Overview:
Built a complete multi-layer neural network from scratch to classify handwritten digits and color images across the MNIST and CIFAR-10 datasets. The project demonstrates a full understanding of the mathematical foundations of deep learning, from activation functions to gradient-based optimization.

Implementation Highlights:

• Coded every component manually using NumPy — including custom ReLU, Sigmoid, and Softmax layers.
• Added dropout regularization to prevent overfitting and improve generalization.
• Implemented stochastic gradient descent (SGD) and momentum optimizers for efficient learning.
• Visualized loss and accuracy curves over training epochs to compare network behaviors.
• Experimented with TensorFlow to validate results and benchmark against higher-level frameworks.

Results:

• Achieved a final test accuracy of 86% on MNIST after tuning learning rate and network depth.
• Improved model stability and convergence speed compared to baseline versions.
• Demonstrated clear performance differences between activation and optimization combinations (Sigmoid + SGD vs ReLU + Momentum).

Insights:
Gained deep intuition for how neural networks learn — understanding vanishing gradients, the effect of activation functions on convergence, and the importance of regularization. The project strengthened both theoretical and practical skills in implementing neural networks from first principles.

Tech Stack:
Python · NumPy · TensorFlow (for validation) · Matplotlib · Jupyter Notebook

Student Performance Prediction (Kaggle)

Mar 2024 – May 2024

Built a regression pipeline to predict students’ grades and explain key drivers of performance. Performed EDA, encoded 17 categorical columns, and used K-Means (k=4 via silhouette) to flag outliers; removing them materially improved every model.

ML&AiReport (3)

Models & Approach

• Trained 7+ models: Linear, Ridge, SVR, Decision Tree, and Neural Networks (basic/improved/advanced).
• Implemented an outlier workflow (distance-to-centroid threshold = mean + c·std) with toggleable removal.

Results (highlights)

• Baseline Ridge: R² ≈ 0.1463 (best among linear models pre-cleaning).ML&AiReport (3)
• Decision Tree: severe overfitting (train perfect fit; test R² ≈ −0.652, EV ≈ −0.642; MAE ≈ 3.76, MSE ≈ 23.83).ML&AiReport (3)
• After outlier removal: all models improved; the Basic Neural Network became top performer (largest practical generalization gains). Improvement percentages: DT +52%, Linear +36.71%, Ridge +32%, SVR +143.86%, Basic NN +135.35%, Improved NN +659.41%, Advanced NN +111.11%.
• Feature insight: parents’ education strongly correlated with outcomes (Ridge coeffs: Medu ≈ 1.1432, Fedu ≈ 0.5773, Health ≈ −0.0601; NN weights echoed Medu/Fedu importance; health negligible).

Tech Stack
Python · pandas · NumPy · scikit-learn · TensorFlow/Keras · Matplotlib

Automated Greenhouse System

May 2021 – Feb 2022

Built a fully automated Arduino-based greenhouse designed to optimize plant growth through precision monitoring and control. Conducted an experimental study comparing automated vs. traditional cultivation using two equal cucumber batches.

Implementation Highlights:

• Developed a control network using Arduino UNO with DHT11 temperature/humidity, soil-moisture, and water-temperature probes, connected through relays to pumps and fans.
• Programmed three closed feedback loops regulating air temperature, humidity, irrigation timing, and water temperature.
• Logged sensor data and system responses in real time to evaluate environmental stability and crop performance.
• Calibrated thresholds (20–38 °C; 40–75 % humidity; soil-moisture ≥ level 3) for autonomous actuation.
• Achieved 88 % higher harvest rate and 50 % improvement in quality yield (90 % Grade-1 cucumbers) compared with manual control.

Social & Technical Impact:
Demonstrated how low-cost open-source microcontrollers can increase agricultural productivity while reducing human error and long-term operating costs. Highlighted automation’s potential role in meeting FAO’s forecasted 70 % rise in food demand by 2050.

Tech Stack:
Arduino UNO · C++/Arduino IDE · DHT11 · Soil Moisture Sensor · Relay Modules · Water Temperature Probe · Electrical Circuit Design

AirVia Airline Booking & Management System (Nexus)
Jan 2023 – Apr 2023
Role: Developer — Backend & GUI IntegrationCollaborated in a six-member team to design and implement Nexus, a full-stack airline ticketing and agency management system for AirVia Travel Agency. The system was developed in Java with a MySQL backend and implemented multi-role access (System Administrator, Office Manager, and Travel Advisor) for managing sales, commissions, and reporting.Implementation Highlights:
Built the database connectivity layer using JDBC and ConfigurationMySQL.java, ensuring secure, modular access to tables for users, sales, blanks, and commission rates.
Developed and maintained backend controller classes such as UserController, SalesController, and BlankController to support CRUD operations and transactional integrity across the MySQL schema.
Implemented authentication logic in UserSL.java to verify user credentials and return persistent session objects, enabling role-based routing across the GUI.
Designed and coded GUI components in Java Swing (JDialog), including dynamic tables for user, blank, and commission management with validation, update, and archive workflows.
Integrated the Apache POI library to generate automated Excel reports (domestic and interline sales) and implemented file export with JFileChooser for user-selected storage.
Contributed to testing and debugging across all user roles, ensuring consistent data flow between GUI, backend, and database through 20+ passed use-case tests (e.g., login, staff creation, blank archiving, sales processing).Team 7_ Nexus Implementation Re…Team 7_ Nexus Implementation Re…Results & Impact:
Delivered a working prototype simulating an enterprise-grade airline management platform capable of handling user authentication, ticket allocation, commission tracking, and financial reporting — all from an intuitive desktop interface. Improved data handling efficiency and reduced error risk by automating core processes previously done manually.Tech Stack:
Java (Swing, JDBC) · MySQL · Apache POI · IntelliJ IDEA · Object-Oriented 

Most Innovative Intern (Eden 2023), Bronze Division 1st place Google Cloud Skill Boost, IYA bronze award and silver award. Google Cloud Skill Boost (cloud computing), Coursera (data science), Harvard CS50, Google code-in (coding challenge participant) & Codeforces (participation).

Interests : Travelling, Tennis, Boxing, padel, Football and Networking

Arduino projects: Smart Temperature & Humidity Monitor, Traffic light safety system, Automated greenhouse

Extra-Curricular Actives: Food packaging and volunteering work NGO (2020-2023) contributed to approximately 150 hours to better welfare and community support. Competitor and Founder at the 2020 INJAZ competition (egypt), seen as the winning team for most innovative product.