About

I'm an NLP Engineer at Xbites, building the AI backend for Darin. I research Arabic embeddings with Hamza Salem Lab and contribute to NAMMA for open-source Arabic AI.

My thesis explores efficient transformer architectures for CPU deployment—focusing on model compression, quantization, and edge AI.

Research Metrics: 7 papers • 54 citations • 962 papers read

Current Focus

Arabic NLP — Building efficient language models for Arabic text understanding
Edge AI — Compressing transformers to run on resource-constrained devices
RAG Systems — Semantic search and retrieval with Qdrant for production
Research Engineering — Bridging academic research with production ML systems

Selected Work

BertHash-Femto

113× smaller than AraBERTv2 with 94% accuracy. Runs inference on edge devices.
GitHub • Python, PyTorch, ONNX

Zarra & Bojji

Tiny Arabic language models optimized for mobile devices.
Article • Model Compression, TensorFlow Lite

كم كالوري (KamCalorie)

Arabic-first nutrition search engine with NLP-powered food recognition.
Live • FastAPI, Astro.js, Arabic NLP

GPUVec

Real-time GPU pricing tracker and ML benchmarking platform.
Live • Data Engineering, Web Scraping

SEO Rat

SEO optimization tool for markdown-based static sites.
GitHub • Python, NLP, Markdown Processing

Technical Background

Languages: Python, Rust, C++, JavaScript
ML/AI: PyTorch, TensorFlow, Transformers, vLLM, FastAI
MLOps: Docker, Podman, CUDA, ONNX Runtime
Web: FastAPI, Astro.js, FastHTML, Svelte, Supabase
Databases: Qdrant, PostgreSQL, Vector Search

Writing

I write about Arabic NLP, model compression, and AI engineering at kareemai.com/blog:

• What is Federated Learning
• Cloud GPU Pricing in 2025
• Building Zarra: Tiny Arabic Models

Subscribe to my newsletter: gpuvec.substack.com

Research

Google Scholar • ResearchGate • Papers

Connect

LinkedIn • X/Twitter • Upwork • Email