DATA PIPELINES · AI SYSTEMS · RESEARCH

Mouad Elbaz

Data Engineer & AI Research Engineer

Building data pipelines, RAG systems, and reproducible ML/LLM experiments.

I build data and AI systems combining scalable ELT pipelines, retrieval-augmented generation, machine learning, and reproducible experimentation. My recent work includes CCSR, Qwen MedQuAD fine-tuning research, and production-style data engineering systems.

Portrait of Mouad Elbaz

STACK

Technologies

Tools I use across data engineering, AI research, backend systems, and cloud infrastructure.

  • Python
  • Apache Airflow
  • dbt
  • Apache Spark
  • FastAPI
  • PostgreSQL
  • Qdrant
  • Docker
  • PyTorch
  • QLoRA

Data Engineering

  • Python
  • SQL
  • PySpark
  • Apache Spark
  • Apache Airflow
  • Celery
  • dbt
  • Hadoop
  • ETL / ELT
  • Medallion Architecture

AI & Machine Learning

  • RAG
  • QLoRA
  • LLM Fine-Tuning
  • Hybrid Retrieval
  • BM25
  • FastEmbed
  • PyTorch
  • Scikit-learn
  • XGBoost
  • Pandas
  • NumPy

Data & Backend

  • PostgreSQL
  • MySQL
  • Redis
  • Qdrant
  • MinIO
  • FAISS
  • FastAPI
  • Spring Boot
  • Next.js
  • REST APIs

DevOps & Cloud

  • Docker
  • Git
  • Linux
  • Oracle Cloud Infrastructure (OCI)

Case studies

Featured Projects

A collection of my most impactful work in data engineering, AI, machine learning, and full-stack development.

Project overviewLocal-first RAG

CCSR — Canonical Computer Science Research

01

AI Research Engineering Platform

CCSR — Canonical Computer Science Research

An open-source, local-first AI research engineering platform evolved from RAGForge. CCSR is designed around reproducible experimentation, evaluation, artifact lineage, retrieval tracing, and research-oriented AI workflows. Its current mature domain is Retrieval-Augmented Generation.

  • Durable Bronze / Silver / Gold ingestion using MinIO
  • PostgreSQL metadata, lineage, and authoritative state
  • Dense + BM25 sparse hybrid retrieval with Qdrant
  • Configurable chunking strategies
  • Retrieval tracing and source inspection
FastAPINext.jsPostgreSQLQdrantMinIORedisAirflowCeleryDockerBM25FastEmbed
View Details — CCSR — Canonical Computer Science Research
  • Durable Bronze / Silver / Gold ingestion using MinIO
  • PostgreSQL metadata, lineage, and authoritative state
  • Dense + BM25 sparse hybrid retrieval with Qdrant
  • Configurable chunking strategies
  • Retrieval tracing and source inspection
  • SSE streaming
  • Redis caching and ingestion events
  • FastAPI control plane
  • Next.js frontend
  • Airflow and Celery ingestion orchestration
  • Benchmarking for latency, throughput, retries, recovery, duplicate processing, and scaling behavior
GitHub
Project overviewFine-tuning research

Qwen MedQuAD — Medical LLM Fine-Tuning Research

02

LLM Fine-Tuning & Evaluation Research

Research in progress

Qwen MedQuAD — Medical LLM Fine-Tuning Research

A reproducible research project studying when medical language models should rely on domain fine-tuning, retrieval, or calibrated abstention. The project uses MedQuAD and Qwen to investigate answer quality, grounding, uncertainty, and robustness.

  • Built a leakage-audited MedQuAD cleaning pipeline
  • 16,357 cleaned QA records
  • 14,344 normalized question groups
  • Group-safe train / validation / test split
  • No detected question-group leakage
Qwen3QLoRAPyTorchTransformersPEFTMedQuADPython
View Details — Qwen MedQuAD — Medical LLM Fine-Tuning Research
  • Built a leakage-audited MedQuAD cleaning pipeline
  • 16,357 cleaned QA records
  • 14,344 normalized question groups
  • Group-safe train / validation / test split
  • No detected question-group leakage
  • Fine-tuned Qwen3-4B using 4-bit QLoRA

Research design · Controlled 2×2 comparison

  • Base Qwen
  • Base Qwen + RAG
  • Fine-tuned Qwen
  • Fine-tuned Qwen + RAG

Planned: complete four-system evaluation, retrieval stress tests, final statistical analysis, and larger Qwen model experiments. Research topics include answer quality, evidence grounding, uncertainty, robustness, and selective answering / abstention.

GitHub
real.screenshots9 views

Dashboard, analytics, ELT, RAG, and operational intelligence views.

03

Data Engineering & Machine Learning

PFE · Professional project

Operational Intelligence Platform

A data and AI platform developed during my PFE internship to consolidate, process, analyze, and query enterprise IT operations data.

  • Consolidated 200,000+ records
  • Sources included GLPI ITSM, OCS Inventory, and security logs
  • Asynchronous ELT pipeline
  • Medallion Architecture
  • Bronze / Silver / Gold analytical layers
AirflowdbtQdrantBM25XGBoostIsolation ForestFastAPINext.js
View Details — Operational Intelligence Platform
  • Consolidated 200,000+ records
  • Sources included GLPI ITSM, OCS Inventory, and security logs
  • Asynchronous ELT pipeline
  • Medallion Architecture
  • Bronze / Silver / Gold analytical layers
  • Apache Airflow orchestration
  • dbt transformations
  • Multilingual RAG system
  • BM25 + dense vector retrieval
  • Reciprocal Rank Fusion
  • Cross-Encoder reranking
  • Qdrant
  • XGBoost ticket triage
  • Isolation Forest hardware anomaly detection
  • FastAPI backend
  • Next.js KPI dashboard
GitHub
real.screenshots1 view

Plant disease detection interface and model preview.

04

Computer Vision & OOD Detection

PlantGuard AI — Plant Disease Detection

A deep learning plant disease classification system combining supervised image classification with out-of-distribution detection.

  • CNN, ResNet, and EfficientNet-B3 benchmarking
  • 38 disease categories
  • 70,000+ PlantVillage images
  • Approximately 99.7% F1-score
  • MSP out-of-distribution detection
PyTorchTorchVisionEfficientNet-B3ResNetCNNGOADStreamlitDocker
View Details — PlantGuard AI — Plant Disease Detection
  • CNN, ResNet, and EfficientNet-B3 benchmarking
  • 38 disease categories
  • 70,000+ PlantVillage images
  • Approximately 99.7% F1-score
  • MSP out-of-distribution detection
  • GOAD out-of-distribution detection
  • Streamlit deployment
  • Dockerized application
GitHub

Research

Research Interests

Interested in reproducible AI research, Retrieval-Augmented Generation, LLM fine-tuning, uncertainty estimation, selective prediction, multimodal AI, and explainable AI.

Current research direction

Brain Region-Aware Explainable Artificial Intelligence for EEG-Based Neurological Disorder Detection

Exploring how spatial brain-region information and temporal EEG features can improve both classification performance and interpretability.

Experience

Professional Experience

Apr 2026 – Jul 2026

Data Engineering & Machine Learning Intern (PFE)

Prefecture d'Agadir Ida-Outanane · Agadir, Morocco

Built ELT, RAG, and ML systems processing 200K+ enterprise records using Airflow, dbt, Qdrant, FastAPI, and Next.js.

Medallion ArchitectureAirflowdbtRAGXGBoostIsolation ForestFastAPINext.js

Apr 2025 – Jun 2025

Developer Intern

Prefecture d'Agadir Ida-Outanane · Agadir, Morocco

Built secure internal systems with RBAC, authentication, session management, audit logging, and network/security controls.

RBACAuthenticationAudit LoggingNetwork Security

Learning path

Education

1

Bachelor in Big Data

Universite Sultan Moulay Slimane · Morocco

2026

2

Technicien Specialise en Reseaux et Systemes

ISTA Agadir · Agadir, Morocco

2023 – 2025

3

Baccalaureat Sciences Physiques

Morocco

2022 – 2023

Certifications

Machine Learning Specialization — DeepLearning.AI
Network Security — Cisco
Developer & Cybersecurity Pathways — Microsoft Learn
Introduction to Data Warehousing — 365 Data Science

Contact

Get In Touch

Get In Touch

Ready to start a conversation? I'm always open to discussing new opportunities, creative projects, data platforms, AI systems, and innovative ideas.

Location: Agadir, Morocco

Phone: +212 751 81 12 01

Email: elbaz.ouad1249@gmail.com

GitHub: moad-cod

Portfolio: mouadelbaz.vercel.app

Let's connect

Contact Me