Hello, I'm
AI Engineer building intelligent systems — from fine-tuned LLMs and RAG pipelines to production ML models and autonomous agents. Grounded in a Data Science foundation of predictive modeling, NLP, and data analytics.
I'm a Data Scientist and AI Engineer with a Bachelor's degree in Data Science & Artificial Intelligence from Irbid National University (GPA 4.0, 2025). Raised and educated in Abu Dhabi, UAE, with continued regional ties. My focus today is on AI engineering — building LLM-powered agents, RAG pipelines, and production machine learning systems.
With hands-on experience across deep learning, NLP, computer vision, and AI automation, I bridge the gap between cutting-edge research and real-world business applications. I thrive on solving complex problems and delivering measurable impact through data.
Irbid National University — Sep 2025 · GPA 4.0
Architecting and deploying end-to-end machine learning solutions. I specialize in creating custom models, from CNNs and RNNs to fine-tuned LLMs, that solve critical business problems and drive tangible results.
Turning raw, messy data into model-ready features — the data cleaning, EDA, and feature engineering pipelines that reliable machine learning and AI systems depend on.
Designing and implementing AI-driven automation with LangGraph and Python — conversational agents, chatbots, and RAG-backed workflows that plug directly into real business processes.
A from-scratch LangGraph research agent — a ReAct loop that searches the web via Tavily, pauses for human approval before saving findings, and keeps cross-session memory. Deployed as a FastAPI SSE streaming API with Docker, Sentry, and LangSmith tracing.
View on GitHubA document Q&A system progressing from naive RAG to an advanced pipeline — semantic + recursive chunking, HyDE, and MultiQueryRetriever over ChromaDB, with LangSmith tracing and RAGAS evaluation.
View on GitHubA walkthrough of LangGraph from basics to production deployment — stateful graphs, agents, state and memory, human-in-the-loop, controllability, and long-term memory, each notebook paired with a line-by-line walkthrough.
View on GitHubA 14-part hands-on tutorial building RAG systems from the ground up — covering indexing, retrieval, generation, multi-query, RAG Fusion, RAPTOR, ColBERT, routing, and query construction.
View on GitHubA hands-on, notebook-by-notebook intro to LangChain — building from a simple agent with tools and short-term memory to advanced multi-agent systems with MCP, and finally production-ready, dynamic agents.
View on GitHubCustom Telegram AI assistant enabling non-technical users to query documents and databases via natural language.
View on GitHubClassified 209,000+ news articles into 41 categories (59% accuracy) using a 1D CNN deep learning model.
View on KaggleAutomated detection system using fine-tuned ResNet50 (91% val accuracy) for medical image classification.
View on KaggleDeveloped a smart recycling system that helps automatically classify waste materials (95% accuracy) using CNN and analyzes consumer behavior.
View on KaggleComprehensive data cleaning, feature engineering, and visualization (Tableau) to uncover trends in movie success factors.
View on KaggleBinary image classifier using transfer learning — fine-tuned VGG16 (with ResNet50 for comparison) on 10,000+ augmented images, reaching 99.3% validation accuracy (loss 0.0267) over 50 epochs with EarlyStopping.
View on KaggleA Convolutional Neural Network in Keras classifying handwritten digits from the 70,000-image MNIST dataset — Conv2D / MaxPooling / Dropout layers trained over 30 epochs with EarlyStopping, reaching 98.76% test accuracy.
View on KaggleML model predicting sports car prices with exceptional accuracy (R² = 0.9926) using Random Forest Regressor on 1,007 records.
View on KaggleEnd-to-end ML project predicting diabetes onset (86% accuracy with XGBoost) from comprehensive diagnostic data.
View on KaggleMulti-class model classifying 14,000+ songs into 11 genres from 17 audio features (danceability, energy, acousticness…). Compared XGBoost, CatBoost, and a Keras neural network — a hard problem due to heavy feature overlap between subjective genres.
View on KaggleEnd-to-end ML pipeline predicting diamond prices from 43,000+ records — feature engineering (a combined size feature, encoded cut/color/clarity) and 3 regression models, with XGBoost topping out at R² = 0.9824 (RMSE 535.65).
View on KaggleLangChain Academy · Jun 2026
LangChain Academy · Apr 2026
Robotna · Sep 2025
IEEE Jordan Section · May 2025
Certiport · May 2025
Correlation One · Mar 2025