👋 Hey there! I'm Ahmed

AI Engineer
& Researcher
from Egypt 🇪🇬

I build resource-efficient AI systems that actually work in the real world — from fine-tuned financial LLMs to smart wheelchairs with Arabic voice control. IEEE co-author. Competition winner. Passionate about AI for everyone. 🚀

See My Work ↓ 📄 Download CV 🐙 GitHub 📡 IEEE Paper ↗

🎓3.62 GPA (Cum Laude)

🥇1st FinAI 2025

🎯90.1% CFA Accuracy

📡IEEE Co-Author

🥇 FinAI 2025

1st

International Competition

📊 CFA Benchmark

90.1%

Matches BloombergGPT

⚡ Memory Saved

75%

via 4-bit Quantization

🙋

About Me

I'm a senior at Zewail City of Science and Technology — Egypt's top STEM university — graduating June 2026 with a 3.62/4.0 GPA (Cum Laude) in Communication & Information Engineering.

My research asks a simple question: how do we make AI work where it's actually needed? During my internship at NUST Pakistan, I helped a team win 1st place at FinAI 2025 (IEEE) by showing a carefully fine-tuned smaller model can beat billion-dollar proprietary systems.

At Takhail, I build production RAG pipelines and real-time multimodal security systems. My graduation project Rafeeq is a smart wheelchair with offline Arabic voice recognition — because healthcare AI needs to speak everyone's language, literally.

Especially excited about Arabic NLP, efficient LLM adaptation, and systems that matter to underrepresented communities. Planning to pursue a Master's in Computer Science next!

💡 "Rafeeq" (رفيق) means "companion" in Arabic — because good AI should feel like a trusted friend, not a machine.

📍

Location

Giza, Egypt

🎓

University

Zewail City of Science & Technology

B.Sc. Comm. & Info. Engineering · Class of 2026

🌐

Languages

Arabic (Native) · English (IELTS B2)

🔍

Open To

AI Research · Applied ML · LLM Systems

⚙️ Tech Stack

Python PyTorch LoRA/QLoRA HuggingFace Unsloth RAG LangChain ChromaDB RF-DETR CLIP 3D-CNN YOLO FastAPI Docker Redis C++ SQL Bash

💼

Experience

AI Engineer

Dec 2025 – Present

Takhail · Giza, Egypt 🟢 Current

Architected a production RAG system for medical consultations using ChromaDB & FastAPI — improved response relevance by 35% and achieved sub-1s latency via Celery + Redis caching.
Designed a multimodal security platform: RF-DETR + CLIP for weapon detection (67% fewer false positives) + 3D-CNNs for violence recognition at 89% accuracy on 16-frame sequences.

AI Research Intern

June – Aug 2025

SEECS Lab, NUST · Islamabad, Pakistan 🏆 IEEE Winner

Part of the team that won 1st Place (Task 2) at FinAI 2025 (IEEE CSCloud) — fine-tuned Llama 3.1 (8B) & Qwen using LoRA/QLoRA on CFA, XBRL, and Bloomberg datasets.
Achieved 90.1% accuracy on CFA Level I (matching BloombergGPT) and 89.6% F1 on financial NER while cutting memory by 75% via 4-bit quantization.
Built modular Unsloth training pipelines + layout-aware OCR for multimodal financial document processing.

Student Representative

2021 – 2025

Zewail City of Science and Technology

Elected representative for my batch (3 years) and all CIE batches (1 year) — liaised with faculty on academic concerns and course logistics for hundreds of students.

Junior Teaching Assistant — Math 101

2022

Zewail City of Science and Technology

Assisted faculty in delivering course content; supported students through tutorials, assignments, and one-on-one guidance.

🚀

Projects

🧠

⌥ GitHub 📄 IEEE

🥇 FinAI 2025 Winner 📡 IEEE CSCloud 2025

FinVerse Agent — Resource-Efficient Financial AI

Modular AI system with 15 domain-specific LoRA adapters on 4-bit quantized LLaMA-3.1-8B. Reduces trainable parameters by 99.74% (7.6B → 20M!) while hitting 90.1% accuracy on CFA Level I. Outperforms GPT-4 and BloombergGPT on 8/10 financial benchmarks with 4× faster inference and 75% lower cost.

LLaMA-3.1-8B LoRA/QLoRA 4-bit Quantization Unsloth Financial NLP Task Routing

🛡️

School Safety System

Two-stage weapon detection: RF-DETR detects + CLIP semantically verifies → 67% fewer false positives. Aggression detection via 3D CNNs. Multi-threaded FastAPI backend with sub-second response across multiple camera feeds.

RF-DETRCLIP 3D-CNNFastAPI

⌥ GitHub 🎬 Demo Video

♿

🎓 Graduation Project 🔧 In Progress

Rafeeq — Smart Wheelchair AI

We are currently building Rafeeq — an intelligent wheelchair companion designed to restore independence for people with mobility challenges. Our goal is to develop an offline Arabic speech recognition system running on Raspberry Pi, making voice control accessible without internet dependency. We aim to integrate real-time health monitoring (HR, SpO₂, Temperature) with automatic emergency alerts via Twilio on fall detection or critical vital signs. The vision: edge AI that speaks Arabic, works anywhere, and genuinely serves underserved communities.

Arabic ASREdge AI Raspberry PiIoT

⌥ GitHub

🏥

AI Medical Consultation Chatbot

Bilingual Arabic/English medical assistant with WhatsApp-style UI. RAG-grounded answers + TTS voice interaction. Production backend: FastAPI + Celery + MongoDB + Redis with persistent history.

RAGArabic NLP FastAPIDocker

⌥ GitHub 🎬 Demo Video

📊

DSTN for CTR Prediction

Deep Spatio-Temporal Neural Networks (KDD 2019) implementation. Dual-stream ConvLSTM + Conv3D architecture capturing spatial ad context and temporal user behavior for click-through rate prediction.

ConvLSTMConv3D PyTorch

⌥ GitHub

🔍

Medical Semantic Search Engine

CPT code classification (87% Top-3 accuracy) using ELMo contextualized embeddings. BiLSTM-based semantic search for sub-second clinical note retrieval — healthcare NLP in practice.

ELMoBiLSTM Medical NLP

📄

Publications

August 2025

📡 IEEE CSCloud 2025 · Special Track: SecureFinAI

FinVerse Agent: Resource-Efficient Financial LLMs with Modular LoRA Fine-Tuning and Task Routing

L. Rana, A. Amin, H. Ameer, and S. Latif

A modular adapter framework for domain adaptation across 15 financial NLP tasks using a 4-bit-quantized LLaMA-3.1-8B base model. Achieves 10× computational reduction while matching or outperforming proprietary models (GPT-4 and BloombergGPT) on financial benchmarks. Demonstrates that parameter-efficient fine-tuning can make LLMs practical in resource-constrained environments.

90.1%

CFA Accuracy

75%

Memory Saved

10×

Compute Reduction

LoRA Adapters

Read on IEEE Xplore ↗

🔬

Research Interests

⚡

Efficient LLM Adaptation

LoRA, QLoRA, and low-bit quantization techniques that make large models viable in low-resource and edge environments — proving you don't need billion-dollar compute to build great AI.

🌍

Arabic & Multilingual NLP

Building language models that genuinely serve Arabic-speaking communities — dedicated pretraining, dialect-aware tokenization, and evaluation frameworks for 400M+ underserved speakers.

👁️

Intelligent Security & Vision

Multi-modal architectures integrating vision, audio, and sensors for real-world automated detection — with rigorous focus on accuracy and false-positive reduction in production deployments.

🏥

Healthcare AI

Real-time monitoring and clinical decision-support systems — vital-sign analysis, offline voice interfaces, and bilingual patient communication for accessible, equitable care.

🏆

Awards & Honors

🥇

1st Place — FinAI 2025

Task 2 winner at IEEE CSCloud 2025 (12th International Conference on Cyber Security & Cloud Computing). Financial AI modeling competition.

⭐

Golden Star Award

Awarded for maintaining GPA above 3.5 for 7 consecutive semesters + 5-Year Academic Excellence Scholarship from Zewail City.

🎯

CFA Benchmark Match

Achieved 90.1% on CFA Level I — matching BloombergGPT — with a model 10× smaller and 75% more memory-efficient.

📡

IEEE Co-Author

Published in IEEE CSCloud 2025 Special Track on SecureFinAI. Available on IEEE Xplore.