About Projects Experience Contact ⬇ Download CV GitHub ↗
AI/ML Engineer

Zeyad
El-Sayed

Building intelligent systems and bringing research into production that actually run in production

|
SCROLL

I build AI that
works in the real world

I'm an AI/ML Engineer who bridges the gap between cutting-edge research and production-grade systems. My focus is on building intelligent systems that are not only accurate but also reliable, efficient, and maintainable under real-world constraints.

My work spans the full lifecycle: designing RAG pipelines and hybrid retrieval systems, fine-tuning large language models with modern PEFT techniques, and engineering multi-modal AI solutions for detection, classification, and generation tasks.

I hold a B.S. in Computers & AI from Benha University, where my graduation project — a multi-modal AI-content detector — achieved over 95% accuracy across image, text, and audio modalities.

10+
Projects shipped & deployed
95%+
Avg accuracy across AI detection tasks
99%
Parameter reduction via QLoRA fine-tuning
3.4
GPA / 4.0 — B.S. Computers & AI

Technical Stack

LLM Engineering
RAG Systems LangChain Fine-Tuning (LoRA / QLoRA) PEFT Prompt Engineering HuggingFace
Deep Learning & CV
PyTorch Transformers CNNs / GANs Keras Scikit-learn OpenCV
Backend & MLOps
FastAPI Docker GitHub Actions Railway Cloudflare Streamlit
Languages & Data
Python C++ SQL Tableau Power BI

Selected Projects

Multi-Modal AI
Catch the AI

Multi-modal AI-generated content detector covering images, text, and audio. Built as a graduation project using state-of-the-art detection architectures, deployed as a full-stack web application.

  • ViT → 96% accuracy on AI-generated images
  • DeBERTa + RoBERTa with an ensemble technique using a feedforward model → 95% accuracy on AI-generated text
  • Wav2Vec 2.0 → 90% accuracy on AI-generated audio
ViTDeBERTaWav2Vec 2.0DjangoReactPostgreSQL
LLM Engineering
Fine-Tuning LLMs

A comprehensive fine-tuning repository covering NLP tasks with deployed models on HuggingFace Spaces. Applies modern PEFT techniques for efficient training with minimal compute.

  • LoRA & QLoRA — up to 99% reduction in trainable parameters
  • Evaluated with ROUGE-L, BLEU — 95%+ performance across tasks
  • Models publicly deployed and accessible on HuggingFace Spaces
PyTorchLoRA / QLoRAPEFTHuggingFace
📝 Classification
📄 Summarization
✍️ Text Generation
🔧 LLMs from Scratch
RAG / Search
Semantic Search
& RAG Guide

A hands-on end-to-end guide to building Retrieval-Augmented Generation systems — from semantic search fundamentals to full RAG pipelines. Designed for practitioners at all levels.

  • Covers semantic search, dense retrieval, and how LLMs integrate with retrieved context
  • End-to-end pipeline: embedding, indexing, retrieval, reranking, and generation
  • Practical resource bridging theory and production implementation
RAGSemantic SearchEmbeddingsLLMsPython
Deep Learning
LLMs from Scratch

A complete implementation of a Large Language Model from scratch in pure PyTorch — built to deeply understand the internals of modern transformer architectures.

  • Tokenization, multi-head self-attention, positional encodings, full transformer blocks
  • End-to-end training & evaluation pipelines built from the ground up
PyTorchTransformersNLP
ML / Data Science
ML Research Projects

Two focused machine learning projects addressing real-world data challenges: class imbalance treatment and credit risk classification — both production-reliability focused.

  • Benchmarked SMOTE, TomekLinks, and hybrid sampling across KNN, XGBoost, RF, SGD, DT classifiers
  • Automated model comparison via PyCaret (Extra Trees, RF, XGBoost) on financial credit data
  • Statistical handling of missing & incorrect values using domain-appropriate strategies
⚖️ Imbalanced Data
💳 Credit Score
PythonScikit-learnPyCaretXGBoostimbalanced-learnTableau
End-to-End ML Dockerized
End-to-End ML Apps

Production-style ML applications demonstrating containerization, serving, and UI — taking trained models from notebook to fully functional deployable apps.

  • FastAPI backend + Streamlit UI + Docker Compose — each project fully containerized
  • Emphasis on deployment pipeline over model training alone
⚽ FIFA Rating Predictor
📩 SMS Spam Detector
FastAPIStreamlitDockerDocker ComposeDNN
Experience & Education

Where I've Worked
& Studied

ITI — Information Technology Institute Jul 2023 – Sep 2023
Business Intelligence Trainee
Egypt · On-site
  • Mastered SQL fundamentals, relational data modeling, and data warehouse design patterns (star/snowflake schemas).
  • Designed and delivered interactive dashboards using Tableau and Power BI, translating raw datasets into actionable business insights.
  • Applied BI tooling to real-world analytics projects covering KPI tracking, trend analysis, and ETL pipelines.

Benha University

B.S. in Computers & Artificial Intelligence

Graduation Project: Catch the AI — multi-modal AI-generated content detector

Oct 2020 – Jun 2024
3.4
GPA / 4.0
Contact

Let's build
something together

I'm open to full-time roles, freelance projects, and research collaborations in AI/ML engineering.