Nicholas Massad

Ai Software Engineer @Levio

Sherbrooke, QC, CA

Explainable AIDeveloped a proof-of-concept explainable AI (XAI) chat model to assist operators in understanding and communicating decisions made by threat assessment AI systems. Responsibilities included- Conducting an in-depth literature review of XAI techniques such as LIME, SHAP, counterfactuals, and anchors, alongside principles of human understanding- Designing and deploying a REST API web server hosted in a Docker container for offline use with open-source LLMs- Performing prompt engineering to optimize the model for limited resources, minimizing hallucinations while delivering accurate and actionable feedback to operators.Transformer-Based Domain Planning:Led the training of open-source language models (LLMs) to generate and assist in planning tasks within specific domains, improving problem-solving efficiency in complex scenarios. Key contributions- Iteratively creating and cleaning domain-specific problem-plan datasets- Training model layers using the Transformers and PyTorch libraries, with PEFT (parameter-efficient fine-tuning) methods- Retraining token embeddings to enhance model performance in targeted applications.CybersecurityCurrently developing an agentic workflow for cybersecurity threat detection, assessment, and evaluation. Tasks include- Evaluating the performance of various machine learning models on attack datasets- Maintaining a REST API and collaborating with front-end and authentication teams to develop a secure and functional application.Technical ExpertiseThese projects involved extensive use of Python, PyTorch, Docker, PDDL, Java, Transformers, LLMs/chatbots, and prompt engineering, and were managed using Agile methodologies.

Université de Sherbrooke

Bachelor of Engineering - BE, Electrical and Electronics Engineering

Cégep de Sherbrooke

DEC, Natural Sciences

Cégep de Sherbrooke

Biotechnology

Université de Sherbrooke

Master of Science - MS, Artificial Intelligence

Université de Sherbrooke

Physics