Rayen Ghali
About
I'm Rayen Ghali, an Applied AI Research Assistant currently based in New Brunswick, Canada. My primary focus areas are Generative AI, Embodied AI, and Computer Vision. What drives me most is architecting solutions, breaking down complex challenges and designing the systems that bring everything together.
I am currently pursuing a Master of Applied Science (M.Sc.A.) at the Université de Moncton, where my thesis research centers on the integration of VLA-based policies with robotic systems.
Projects
Portfolio Agent
Next.js portfolio with a FastAPI backend agent that controls the UI via accessibility tree snapshots.
Master's Thesis
End-to-End Multimodal AI Agent for Autonomous Robotic Control
ChunkCanvas
Personal project — a multimodal processing GUI for parsing documents into chunked data for vector databases.
RobotVoiceControl
LLM-driven Agent for Industrial Robot Control. Controls a KUKA robot via natural language (text & speech).
ChatPatrimoineAcadien
Web app for acadian archival image search.
ChatCapitalHumain
Multi-agent data analytics and visualization platform for educational survey data.
CrowdKPI
Cross-platform audience analytics app using React Native, FastAPI, YOLOv11 and ViT for emotion recognition.
ChatAcadien
RAG-based conversational agent for querying historical Acadian genealogical records (1700-1900).
Experience
Research Assistant
- Over the past two years, I have contributed to applied AI research projects across agentic systems, robotics, and computer vision working alongside lab members and domain experts, with several projects resulting in deployed applications or peer-reviewed publications.
- Collaborated with the CFRIA lab to design and build a LangGraph-based multimodal agent for controlling an industrial KUKA robot via natural language (text & speech), with persistent memory, real-time error recovery, and sub-3s end-to-end latency. Co-authored 'LLM-driven agent for speech-enabled control of industrial robots: A case study in snow-crab quality inspection', published in Results in Engineering, 2025.
- Developed ChatAcadien, a RAG-based conversational agent for exploring 200 years of Acadian genealogical records, with entity-centric chunking to handle ambiguous historical names and reduce hallucinations. Also built a companion image search chatbot using VLM-generated descriptions (chatpatrimoineacadien.ca). Co-authored and presented 'ChatAcadien: A RAG-LLM-Based Chatbot for Exploring Acadian Genealogy' at IEEE CASCON 2025.
- Built ChatCapitalHumain, a multi-agent analytics platform for exploring 15 years of Acadian educational survey data from 7 schools through natural language queries and automated visualizations.
- Architected and implemented SSL-YOLO, a few-shot industrial defect detection framework combining contrastive self-supervised pre-training with YOLOv8, for settings with very limited labeled data. Co-authored and presented 'Real-time defect detection systems for steel and wood inspection' at IEEE CCECE 2024 and 'Benchmarking few-shot learning techniques for steel surface defect detection' at IEEE SWC 2025.
Data Science Intern
- Contributed to the design and development of a tourist recommendation system for historical places in Tunisia, targeting personalized user experiences based on preference data.
- Applied SMOTE and feature encoding strategies to handle class imbalance and augment mixed numeric and categorical datasets, improving training data quality.
- Benchmarked Content-Based, Collaborative Filtering, and Hybrid recommendation approaches, selecting Collaborative Filtering for its superior performance on sparse user-item interaction data.
- Tech Stack: Python, TensorFlow, Scikit-learn
Backend Developer Intern
- Conducted a comparative analysis of open-source IAM solutions based on the OAuth 2.0 protocol, evaluating their fit for identity management and access control.
- Designed and implemented a Keycloak authentication & authorization layer exposed as a RESTful API within a microservices architecture, enabling RBAC and SSO capabilities.
- Tech Stack: Java, Spring Boot, Angular, PostgreSQL
Education
Master of Applied Science (MASc)
CGPA: 4.15/4.3.
Thesis: "End-to-End Multimodal AI Agent for Autonomous Robotic Control".
Main Courses: Optimization Methods in Engineering, Numerical Methods in Engineering, Engineering Research Seminar.
Industrial and Systems Engineering
Graduated with High Honours.
Main Courses: Algorithmics & Data Structures, Object-Oriented Programming, Probability & Statistics, Control Systems & Industrial Automation, Operations Research, Embedded & Real-time Systems, Computer Vision, Industrial Information Systems (ERP/SAP), Production & Maintenance Management.
Baccalaureate in Mathematics
Graduated with Highest Honours.
Published Papers
R. Ghali and S. A. Selouani, "ChatAcadien: A RAG-LLM-Based Chatbot for Exploring Acadian Genealogy," in 2025 IEEE International Conference on Collaborative Advances in Software and Computing (CASCON), Toronto, ON, Canada, 2025.
I. Kadri, S. A. Selouani, M. Ghribi, R. Ghali, and S. Mekhoukh, "LLM-driven agent for speech-enabled control of industrial robots: A case study in snow-crab quality inspection," in Results in Engineering, vol. 27, p. 106660, 2025.
R. Ghali, Z. Benhafid, and S. A. Selouani, "Benchmarking Few-Shot Learning Techniques for Steel Surface Defect Detection," in 2025 IEEE Smart World Congress (SWC), Calgary, AB, Canada, 2025.
R. Ghali, Z. Benhafid, and S. A. Selouani, "Real-time defect detection systems for steel and wood inspection," in 2024 IEEE Canadian Conference on Electrical and Computer Engineering (CCECE), Kingston, ON, Canada, pp. 577-582, 2024.
Certificates
Scholarships & Fellowships
Mitacs Globalink Research Internship
Mitacs Business Strategy Internship (BSI)
Mitacs Globalink Graduate Fellowship
New Brunswick Graduate Scholarships (BESNB)
Extracurricular Activities