Moaz Eldegwy: Hero

I’m Moaz, an AI engineer.

I train language models to reason and act.

LLM post-training, agentic systems and retrieval, from dataset to deployment.

Selected work Résumé
Based in Giza, Egypt Currently Open to AI engineering roles
Moaz

I’ve always wanted to understand how things work, then explain them, then build them better. At eleven it was circuits and robots. Today it’s language models that learn, reason and act.

Electronics at 11 · freelancing at 16 · AI since 2023

Selected work

AI systems, from research models to the products around them.

All projects

Research questions

I turned to research in my third year and made my graduation project a research project.

  • Can small language models learn to correct themselves over many steps?

    Graduation research · 2025–26

    LABD: distilled agentic loops, SFT and GRPO to get small models past the “Reasoning Cliff”. Code, datasets and models are open.

  • Can an agent reinforce itself through supervised self-distillation?

    In progress

    With Ahmed Elemam, in Google DeepMind’s Gemma 4 Developer Agent Competition, running until December 2026.

Interests LLM post-training · RL for language models · Agentic systems · Retrieval

Skill map

Every project is linked to the skills and tools used to build it. Hover or tap a project, a skill or an area to trace its links.

All skills
  • LLMs Dataset Engineering, Generative AI, Hugging Face, Knowledge Distillation, Large Language Models (LLM), LLM Fine-Tuning, LoRA, Model Evaluation, Quantization, Supervised Fine-Tuning (SFT)
  • Reinforcement learning Reinforcement Learning, Reinforcement Learning (GRPO)
  • Agents Agentic AI, LangGraph, Multi-agent Systems
  • Retrieval Knowledge Graphs, Retrieval-Augmented Generation (RAG), Vector Search
  • Tools Google Cloud, Python, PyTorch, Weights & Biases

From circuits to language models

Eleven years of building, leading and competing, newest first. Each entry sits in the year it started.

2026

2025

2024

2023

2022

2020

2017

2015

  • 2015 – 2019 Electronics, robotics & smart homes Self-taught at 11 · robots and custom PCBs Projects
Moaz Eldegwy: Speaking

I teach what I learn,
and build teams that outlast me.

I lead with an eye on impact, building every team I work with into a system that keeps evolving long after my term ends.

  • Sep 2026
    LLM Engineering Bootcamp Led a two-day bootcamp for 300+ attendees.
  • Jul 2026
    Generative AI instructor SBS summer camp · four hands-on sessions
  • 2023 – 26
    Microsoft Student Club, AOU Co-founder, AI team lead, then president of 50+ members
  • Talks
    Introduction to LLMs · AOU Research Day 2026 Workshops and research days for students
Talks & events LinkedIn

On video

Sessions where I build and explain AI systems end to end.

YouTube

Building a reasoning model for coding تطبيق عملي

Ahmed Elemam - أحمد الإمام · Jun 2026

With Ahmed Elemam on his channel: a hands-on walkthrough of building a reasoning model for coding. Placeholder summary: rewrite in your own words.

Moaz Eldegwy: Suit

More than one thing, on purpose.

I started freelancing at 16 and delivered more than sixty software projects before moving fully into AI.

As a kid it was robots and circuit boards, and second place at FEE Robotics in 2017. Then came web products, marketing and the servers that ran them. Today it’s language models.

Beyond AI

None of it is the job; all of it shapes how I do it.

  • Since 2015 Robotics & electronics

    Robots, smart homes and custom PCBs. 2nd place, FEE Robotics 2017.

  • Always Math, physics & space

    Aerospace, flight and how the universe works.

  • 2020 – 25 Web & systems

    Freelance engineering, Linux servers and a hosting business.

  • Business Management & marketing

    Leading teams and projects, and taking what I build to market.