Tech4 Student at Epitech | AI Builder & ML Engineer
I am a developer deeply passionate about the state-of-the-art in Artificial Intelligence, LLM engineering, and machine learning fundamentals. My philosophy is simple: understand the theory by building things from scratch, and apply these concepts to optimize inference, design complex architectures, and create autonomous agents.
I document my learnings, experiments, and "build in public" journey on my YouTube channel.
- LLM Optimization & Inference: Model quantization, GGUF conversion, and optimizing local GPU VRAM usage.
- Deep Learning Architectures: Rebuilding core mechanics from scratch, experimenting with Mixture of Experts (MoE), and Spiking Neural Networks (SNN).
- Agentic Workflows & RL: Implementing Reinforcement Learning (PPO) for autonomous agents and orchestrating local LLMs (Ollama) with complex RAG pipelines.
- llm-optimizer A CLI pipeline to quantize any HuggingFace model for local GPU inference. Features automatic VRAM detection, GGUF conversion, and
lm-evalbenchmarking. - MINI_MOE_DEMO An experimental implementation of the Mixture of Experts (MoE) concept, providing clear visualization of routing and expert activation within artificial neural networks.
- PPO-Racetrack-Curriculum-Reinforcement-Learning An autonomous racing agent that learns to drive from zero using Proximal Policy Optimization (PPO) and curriculum learning.
- my_neural_network My own neural network library built entirely from scratch using NumPy. A fundamental project to deeply master mathematical mechanics and backpropagation without relying on standard frameworks.
- physlint A lightweight latent-space model (SVD + ridge regression) that learns what "normal" video looks like from a training set to automatically detect anomalies in new videos.
I'm always open to discussing AI architectures, inference optimization, or my latest experiments.
📧 amdjad.freelance@gmail.com
🔗 LinkedIn
