Open to PhD Positions · Fall 2026

Kareem Waly

Machine Learning Engineer & Published AI Researcher (IEEE)

|

I build production Arabic AI systems — and I'm seeking a PhD where that work becomes research. My work spans Graph Neural Networks for drug discovery, multimodal Arabic-English sentiment AI, and retrieval-augmented systems. Published at IEEE ASET 2025. Technical Program Chair, IndabaX Egypt 2026. Co-founder of AI-Talents (33 mentees).

View Systems View Research CV
For consulting, research discussion, or PhD opportunities
Kareem Waly
0
IEEE Publication
0
Live HF Spaces
0
Mentees · AI-Talents
A+
M.Eng · uOttawa

My Story

From applied mathematics to peer-reviewed AI — the path that brought me here.

My journey didn't start by planning to become an AI researcher, it started by wanting to understand things deeply. To find my why. And AI kept being where the hardest, most interesting questions lived.

That curiosity took me from mathematics and computer science at Helwan University to a full scholarship M.Eng at the University of Ottawa, to publishing IEEE research at ASET 2025, to building production AI systems that real businesses depend on today. Each step wasn't a credential, it was a question I needed to answer.

The question I keep coming back to is this: why does technology still struggle to understand people in their own language, in their own context, in the way they actually live? Egyptian Arabic alone is spoken by nearly 110 million people. Arabic speakers globally number around 500 million native and non-native. And yet the models shaping their digital world in healthcare, in education, in daily life were trained almost entirely on someone else's language, someone else's data, someone else's assumptions. That's not just a technical gap. It's a representation gap, and it has real consequences for real people.

I build toward closing it through production systems, through research, through community. Whether that means a better diagnostic tool that works in Arabic, a speech model that understands how patients actually speak, or a language system built for the people it serves rather than around them. I'm looking for a PhD where that work can go deeper, get published, and reach further than I can reach alone.

ML Engineer · Published IEEE Researcher · Bridging Rigour & Impact 🧠⚗️🎓

Research-First Approach

From concept to peer-reviewed IEEE publication — proven end-to-end research capability.

Full ML Lifecycle

Preprocessing to deployment and monitoring. Spoken Language ID system with 90%+ accuracy in production.

Educator & Leader

Co-founded AI-Talents (33 mentees). Technical Program Chair at IndabaX Egypt 2026.

Cross-Cultural Collaborator

Egypt, Canada, Germany, Switzerland, UAE. Adaptable, open-minded, detail-oriented.

01 · Selected Systems

Four systems, each answers why · what · matters.

Not a projects gallery. These are the systems I stand behind — research-grade or production-ready, each with a clear reason to exist.

Flagship · Drug Safety

Graph Neural Networks for Drug–Drug Interaction Prediction

Two-layer GNN over a shared molecular encoder. Layer 1 predicts the DrugBank interaction type (86 classes). Layer 2 predicts TwoSides polypharmacy side effects (multi-label). Published in IEEE Xplore.

Why It Exists
Most DDI systems answer only one clinical question and validate only on warm splits that mask novel-drug generalization.
What It Does
Edge-aware GINEConv encoder (4 layers, weight-tied) + bilinear co-attention head for relation type; same encoder frozen + multi-label sigmoid head for side effects.
Why It Matters
Two complementary clinical questions answered by one model. Honest warm + cold-drug evaluation. Reproducible pipeline with open weights and a live demo.
96.39 % top-1 warm99.69 % top-364.14 % top-1 cold-drug0.786 macro-AUROC
PyTorch GeometricGINEConvRDKitDrugBankTwoSidesFastAPIDocker
↗ Live Demo 🔒 Request Code Access
Flagship · Multimodal

Multimodal Sentiment & Emotion System

Three-stream fusion (text · audio · video) that reads sentiment and emotion across English and Arabic video input — in real time.

Why It Exists
Text-only sentiment analysis is flat — it misses sarcasm, tone, and facial contradiction that multimodal signals capture.
What It Does
Fuses DeBERTa text, MFCC voice, and ArcFace video representations. Arabic cross-lingual transfer via AraBERT.
Why It Matters
80.06% accuracy across 7 emotions; works across languages; deployable as a real-time Gradio interface.
80.06% acc71.49M params7 emotions
PyTorchDeBERTaAraBERTArcFace
↗ Try All 3 Demos 🔒 Request Code Access
Flagship · Healthcare

Brain Tumor MRI Classifier

A CNN that triages MRI slices for tumor presence with clinician-readable confidence — deployed as a Gradio space for non-technical validation.

Why It Exists
Radiology backlogs delay diagnosis — an automated first-pass helps prioritize urgent cases.
What It Does
Transfer-learned CNN on curated MRI data. End-to-end: image reading → classification → confidence output with Grad-CAM visualization.
Why It Matters
Accessible to non-ML radiologists via a web interface; a technology can accelerate, not replace, clinical review.
Transfer LearningGrad-CAM
MRI classificationGrad-CAMLive Demo
↗ Live Demo 🔒 Request Code Access
Flagship · RAG

Hybrid RAG Pipeline for Log Forensics

A five-stage dense+sparse retrieval pipeline that turns 2M unstructured logs into grounded incident reports.

Why It Exists
Log analysis at scale is manual and error-prone; traditional search fails on unstructured, noisy operational data.
What It Does
BM25 + BGE hybrid retrieval, SBERT re-ranking, Qwen2.5-7B generation. Five pipeline stages from ingestion to report.
Why It Matters
+18% MAP over dense-only; generates grounded, citation-backed incident reports from raw logs.
+18% MAP2M log linesQwen2.5-7B
BM25BGESBERTQwen2.5
↗ Live Demo 🔒 Request Code Access

Secondary tier — shipped work, strong signal.

Below the flagships: five applied systems that round out the engineering breadth — agentic AI, distributed systems, AI speech, biometrics, security.

Applied· Agentic

AI Agent · Customer Support & Sales Analytics

Enterprise agentic workflow with CRM integration, automated support, and operational dashboards.

Agentic AICRMAnalyticsWorkflow
Private · implementation in progress
Applied· Systems AI

AI-Driven Dynamic Modeling · Distributed DBs

Three-phase AI framework for fragmentation (Random Forest 97.88%), replication (Q-Learning), and anomaly detection (LSTM autoencoder).

Random ForestQ-LearningLSTMDistributed
Deployed · infra-AI systems bridge
Applied· Speech

Spoken Language Identification

Production-deployed language-ID pipeline at AIM Technologies — 90%+ accuracy across the full ML lifecycle.

SpeechAudio ClassificationProduction
Shipped in industry · case study pending
Applied· Biometrics

Multimodal Biometric Authentication

Fused face + voice + fingerprint signals for robust identity verification under noise and spoofing.

FaceVoiceFingerprintFusion
Archived · systems-integration proof
Applied· Security

Next-Gen Intrusion Detection (GANs)

GAN-based generative modeling for anomaly-resistant intrusion detection on imbalanced attack distributions.

GANsAnomaly DetectionCybersecurity
Archived · cross-domain ML breadth

Publications

Peer-reviewed research.

Predicting Drug-Drug Interactions Using Graph Neural Networks: Towards Safer and Intelligent Prescriptions

Elsersy M., Waly K., Sherif A.

Proceedings of the 7th HCT ASET Conference · Abu Dhabi, UAE · 15–18 September 2025 · IEEE Xplore

Published Author & Presenter · ASET 2025 Applications in Electrical Engineering Track
Google Scholar IEEE Xplore

Experience

Where research meets impact.

Technical Program Chair — IndabaX Egypt

2026 – Present

Lead the technical program for Egypt's largest AI/ML community conference. Curate the workshop track, recruit speakers and mentors, and shape the scientific direction.

Co-Founder & AI Researcher — AI-Talents EdTech

Oct 2023 – Present

Co-founded a mindset-first EdTech startup training the next generation of AI engineers across MENA. Architected the full curriculum. 33 active mentees.

Machine Learning Engineer — AIM Technologies

Aug 2022 – Feb 2023

Shipped production Spoken Language Identification with 90%+ accuracy — full ML lifecycle from preprocessing through deployment and monitoring.

Data Scientist Intern — Hertie School Data Science Lab

Aug – Sep 2023 · Remote, Germany

Researched GANs for synthetic data generation. Contributed to ML pipeline design alongside senior researchers. Expanded training-set coverage by 35% via GAN-based synthetic generation pipeline.

Research Interests

Three coherent directions — each backed by shipped work.

Arabic & Low-Resource NLP

Egyptian Arabic language models, dialectal NLP, speech processing, and multilingual AI for underserved MENA languages. Applied in production retail systems and multimodal sentiment pipelines.

Multimodal AI & Speech

Fusion of text, audio, and visual signals for cross-lingual Arabic/English sentiment and emotion recognition. Spoken language identification deployed at 90%+ accuracy in production.

Healthcare AI & Graph Neural Networks

ML for drug-drug interaction prediction, clinical decision support, and molecular property prediction. Published IEEE work in GNN-based biomedical AI at ASET 2025.

Get in Touch

Research, PhD, or applied AI — I'd be glad to connect.

Whether you want to discuss a PhD opportunity, propose a research collaboration, or scope an applied AI engagement — I'd be glad to connect. For consulting, research discussion, or PhD opportunities, email me directly or book a short call.

Email Me View Research CV
🎓 Research / PhD discussion
💼 Freelance / consulting inquiry
Emailkw.kareemwaly@gmail.com
LinkedInkareemindata
GitHubkareemindata
ScholarPublications

Ways I Collaborate

Research collaboration, PhD discussion, consulting — same person, different tracks.

I work with both research teams and applied clients. My work sits between rigorous experimentation and production-ready AI systems — so I can contribute to research-oriented collaboration, consulting engagements, or full ML system development.

For Research & Academia

Labs · PhD programs · Joint research

Research Collaboration

Literature reviews, experiment design, baseline development, benchmarking, and model prototyping across multimodal AI, GNNs, RAG, and Arabic NLP.

PhD / Research Discussion

Open to PhD opportunities, research assistantships, and interdisciplinary collaborations where machine learning can contribute to meaningful scientific or societal impact.

For Applied AI & Consulting

Companies · Teams · Product

ML Engineering

Production-minded ML systems — from preprocessing and model training to deployment, monitoring, and iteration.

AI Agents & RAG Systems

Retrieval-augmented generation, tool-calling agents, and LLM-powered workflow automation for support, analytics, and internal knowledge systems.

Data Science & Applied Research

Exploratory analysis, modeling, reporting, and research-to-product translation for organizations that need both technical depth and practical delivery.

Looking for a research collaborator or an applied AI partner?

Email me directly or book a short call — choose Research / PhD or Consulting as the meeting type.

Email Me

Community & Volunteering

Growing the AI ecosystem in MENA and beyond.

IndabaX Egypt 2026

Technical Program Chair

AI-Talents EdTech

Co-Founder · 33 mentees

Curriculum

IEEE ASET 2025

Published Author & Presenter

Certificate

DDiB 2025 · UZH Zürich

Ambassador — Deep Dive into Blockchain

Ambassadors page

Skills

Research to production.

Research

GNNsArabic NLPMultimodal AIRAGAI AgentsGen AIRLAnomaly Detection

Languages & Frameworks

PythonPyTorchTensorFlowHugging FaceFastAPIScikit-learnSQL

Infrastructure

GitDockerAWS SageMakerMLOpsVector DBsGradio

About Kareem Waly

Kareem Waly (also Karim Waly, Karim Khaled, kareemindata) is a Machine Learning Engineer and IEEE-published AI researcher. He holds an A+ M.Eng from the University of Ottawa under a full DEBI scholarship, a B.Sc. in Computer Science from Helwan University (2nd in class), and completed Deep Dive into Blockchain at the University of Zurich (UZH).

His research focuses on Arabic Natural Language Processing, Multimodal AI, Graph Neural Networks for Drug–Drug Interaction prediction, Retrieval-Augmented Generation (RAG), multimodal sentiment and emotion recognition, large language models, low-resource NLP, multilingual language models, and applied AI systems for the MENA region.

Published at IEEE ASET 2025: "Predicting Drug–Drug Interactions Using Graph Neural Networks" (IEEE Xplore document 11428029). Co-founder of AI-Talents EdTech (33 mentees). Technical Program Chair at IndabaX Egypt 2026. DDiB 2025 Ambassador at UZH Zürich.

He is currently seeking a fully funded PhD position in machine learning, multimodal speech, low-resource NLP, or multilingual language models — open to research assistant and applied AI research collaborations worldwide.

Technical stack: PyTorch, PyTorch Geometric, DeBERTa, AraBERT, Whisper, RDKit, FastAPI, Docker, Gradio. Locations: Cairo Egypt, Ottawa Canada, Zurich Switzerland.