</>ME
Open to ML & AI Roles

Building intelligent systems, from research to production.

Computer Science & AI student at Al Alamein International University specializing in NLP pipelines, parameter-efficient fine-tuning (LoRA), knowledge distillation, and graph routing optimization.

Mohamed Islam Elshourbagy
Alexandria, EgyptArabic • EnglishCS & AI @ AIU (2027)
// 01 — About Me

Turning Research into Deployed Intelligence

I'm a Computer Science & AI student at Al Alamein International University, graduating in 2027, focused on building AI systems that actually run — not just models in a notebook, but the full path from data to something deployed and usable.

My work spans three areas: parameter-efficient NLP (fine-tuning and distillation), full-stack ML applications (traffic prediction, routing, containerized deployment), and multi-agent LLM systems (agents that research, summarize, and verify each other's work). I'm currently building on this through DEPI, Egypt's national program for agentic and generative AI development.

What I care about most is the last step most student projects skip — turning a working model into something someone else can open, use, and trust. The details and results for each project are below.

BSc Computer Science — Artificial Intelligence

Al Alamein International University (AIU) • Egypt

Oct 2022 — Present

Specialization in Artificial Intelligence covering machine learning, deep learning, NLP, algorithms, and software engineering.

// 02 — Experience

Where I've worked

Agentic AI & Generative AI System Developer

Digital Egypt Pioneers Initiative (DEPI)

Egypt
Jul 2026 - Present
Hybrid
  • Selected for DEPI, a government-backed national technical training initiative focused on building advanced digital skills for the Egyptian tech workforce.
  • Currently developing hands-on expertise in agentic AI system architecture, LLM-based application design, and generative AI development workflows.
  • Engaging with practical training covering AI agent orchestration, prompt engineering, and applied machine learning concepts as part of an intensive, project-based curriculum.
  • Building foundational proficiency in tools and frameworks relevant to generative AI development (e.g., LangChain, RAG pipelines, LLMs).
Agentic AIGenerative AILangChainRAG PipelinesLLMsPrompt Engineering

Game Development & Animation Intern

City of Scientific Research and Technological Applications (SRTA-City)

Egypt
Jul 2025 - Sep 2025
Hybrid
  • Developed 3D character models with complete skeletal rigging in Blender, producing fluid animations and realistic movement for in-game characters.
  • Implemented AI agent behaviors using Unity’s NavMesh system for intelligent pathfinding and navigation, creating NPCs that dynamically respond to environmental obstacles and player actions.
  • Applied character rigging, AI agent behavior programming, and animation pipeline workflows for interactive game development in Unity.
  • Built and textured 3D assets using advanced modeling techniques, ensuring performance-optimized assets suitable for real-time rendering.
BlenderUnityNavMeshC#

Software Development Trainee

Information Technology Institute (ITI)

Egypt
Jul 2024 - Aug 2024
Hybrid
  • Developed a full-featured Chat Room Application in Java (JavaFX, FXML, Swing) with user authentication, real-time messaging, and active status indicators.
  • Implemented multi-threaded server architecture using Java socket programming to handle simultaneous user connections, improving system reliability and performance.
  • Designed responsive and intuitive user interfaces with JavaFX and FXML, applying UI/UX best practices to enhance the user experience.
  • Applied concurrent programming and multi-threading concepts to support real-time data processing across connected clients.
JavaJavaFXFXMLSwingSocket Programming
// 05 — Technical Skills

Tools, Frameworks & Concepts

AI, NLP & Deep Learning

Specialized focus in natural language processing, transformer architectures, parameter-efficient fine-tuning, and neural knowledge distillation.

Natural Language Processing (NLP)Large Language Models (LLMs)Parameter-Efficient Fine-Tuning (LoRA / PEFT)Knowledge Distillation (Teacher-Student)Hybrid Attention PoolingFocal Loss & Label SmoothingDeep Neural Networks (CNN/LSTM)Multi-Agent AI Systems (LangChain)Mixed Precision (FP16/BF16)
9 technologies7 Core Proficiencies

ML Frameworks & Libraries

Production and research frameworks for deep learning, scientific computing, and model evaluation.

PyTorchHuggingFace Transformers & PEFTScikit-LearnTensorFlow / KerasNumPy & PandasMatplotlib & SeabornNLTK & spaCyWeights & Biases (W&B)
8 technologies4 Core Proficiencies

Algorithms & Backend Systems

Graph routing, concurrent network architectures, and containerized backend microservices.

Graph Algorithms (A*, Dijkstra)Dynamic ProgrammingRESTful API Development (Flask)Multi-Threading & ConcurrencySocket Programming (TCP/IP)Docker ContainerizationObject-Oriented Design (OOP)Data Structures & System Design
8 technologies7 Core Proficiencies

Programming Languages

Core languages for AI pipelines, concurrent systems, desktop GUIs, and simulation.

PythonJavaC#SQL / Relational DBsJavaScript / TypeScriptHTML & CSS / TailwindR
7 technologies2 Core Proficiencies

AI Simulation, 3D & Tools

Game AI navigation engines, 3D animation modeling, and developer tooling.

Unity & NavMesh PathfindingBlender Skeletal Rigging & IKJavaFX & FXML GUIGit & GitHub WorkflowsLinux / Bash ScriptingVS Code & PyCharm / IntelliJ
6 technologies4 Core Proficiencies
// 06 — Certifications

Courses & credentials

Neural Networks and Deep Learning

DeepLearning.AI•Coursera

NLP with Classification and Vector Spaces

DeepLearning.AI•Coursera

NLP in TensorFlow

DeepLearning.AI•Coursera

Building Multi-Agent Systems

Microsoft•Coursera

Mohamed Elshourbagy

CS & AI Student • ML Engineer

© 2026 Mohamed Islam Elshourbagy. All rights reserved.

// 03 — Featured Work

Engineered Systems & Pipelines

Full-Stack & Graph Optimization Featured

Intelligent Cairo Transportation System

ML-Powered Route Optimization & Traffic Planning

Containerized full-stack web application combining Scikit-Learn traffic prediction models with A*, Dijkstra, and Dynamic Programming algorithms to simulate adaptive routing across Greater Cairo.

Calculation Latency< 120ms
Algorithms SupportedA*, Dijkstra, DP
Backend ArchitectureFlask REST + Docker
VisualizationInteractive Leaflet
PythonScikit-LearnFlaskDockerLeaflet.jsNumPy
CodeLive Demo
NLP & LLM Optimization Featured

Efficient Sentiment Classification

GPT-2, LoRA & Knowledge Distillation

Parameter-efficient NLP pipeline combining LoRA adapter matrices with teacher-student knowledge distillation, training only 2.15% parameters while retaining 96.9% of full teacher F1 performance.

Param Reduction97.85%
Macro F1 (IMDB)0.8962
Trainable Params2.6M / 124M
Pooling Mechanism80% Attn / 20% Mean
PythonPyTorchHuggingFace TransformersLoRA (PEFT)Weights & Biases
CodeLive Demo
Multi-Agent AI & Verification Featured

Multi-Agent Research Assistant

Autonomous Multi-Agent Orchestration & Source Verification

A multi-agent AI system where specialized agents (Search, Summarizer, Fact-Checker) collaborate under a Coordinator to research a question, synthesize a cited answer, and fact-check it — producing a grounded, source-verified report. The Fact-Checker agent caught and flagged an unsupported statistic during a live test run, demonstrating real self-verification.

Architecture4 Specialized Agents
Search BackendDuckDuckGo (ddgs)
Inference EngineGroq Cloud
Self-AuditAutomated Fact-Check
PythonLangChainGroq (LLM inference)DuckDuckGo Search (ddgs)Streamlit
CodeLive Demo
// 04 — Interactive Lab

Live Model & Systems Playground

Inference Input & Attention Pooling
DistilGPT-2 + LoRA (PEFT)
14 tokens

Hybrid 80% Attention Pooling Heatmap

Higher saturation = higher weight
Thisparameter-efficient90%LoRAdistillationpipelinedrastically85%reducesmemoryfootprintwhilemaintaining87%exceptional89%F1accuracy.
Classification Output
Very Positive

Fine-Grained SST-5 Distribution:

Very Positive86%
Positive11%
Neutral2%
Negative1%
Very Negative0%
Teacher (GPT-2)124.4M params38ms Latency
LoRA Student2.68M trainable14ms (2.7x speed)
// 07 — Get In Touch

Let's Build Something Intelligent

Quick Message Presets:

Direct Communication

Prefer direct channels? Reach out anytime via email, phone, or professional networks. 1-click copy available for quick outreach.

  • info.moislam@gmail.com
  • +20 128 080 6343
  • LinkedIn Profile
  • GitHub Profile
Alexandria, Egypt (Available Globally)