Basel Mousi Email & Phone Number

Atlanta, Ga, Us

Doha, Qatar

Working with the Arabic language technologies group on explainable ai

Project: Preventing Online violence against children using NLP. Tasks: - Data annotation: annotated conversational data by breaking it into sentence instances and then labeling each sentence for whether it is risky, relatively risky, or not risky.- Model Training: Finetuned BERT for sequence classification on the annotated data and reported the results for the team.

Project: AI assisted collaborative mapping tool for disaster management. I was the team lead of the model training part. The goal of the task was to train a model that segments buildings from aerial imagery. The task included the following subtaks: - Training a baseline encoder-decoder CNN model on the given data - Training a UNET model with a ResNet backbone on the given data- Investigation of using an ensemble of UNETS on the data (open cities ai challenge)- Utilizing transfer learning to finetune a pretrained UNET model with an efficientNet encoder on the data The team delivered robust models that met the needed requirements for each of the given regions.

Doha, Qatar

Worked on explainable AI for deep language models as a part of the Arabic language technologies group.

- Developed a Question-Answering chatbot that can be used in OSCE exams and delivered a script in which the medical students at the AUB school of medicine can interact with.- Trained several text binary classification models (sentiment analysis mainly) and studied their explainability using LIME and SHAP models.- Developed a speech-in speech-out emotionally aware chatbot that takes into account the tone of theuser to generate an emotionally aligned speech response- Developed a few-shot multitask learning scheme that can be used to generate text in low resourcesettings.- Designed a chat engine for a mental health chatbot that uses cognitive behavioral therapy to generateits responses.- Designed a chat engine for a smoking cessation chatbot that uses motivational interviewing therapy togenerate its responses

Virginia, United States

Supervisor: Prof. Sara Lu Riggs- Conducted a literature review on virtual reality technologies and on modeling human performance in3D virtual environments (Fitts Law)- Delivered a detailed analysis about factors that contribute to movement time in 3D virtual environments- Built and evaluated several machine learning models that can predict the movement time in 3D virtualenvironments

Beirut, Lebanon

- Created a large dataset that can be used for make, model, year, and color classification of vehicles- Used object detection to detect vehicles on roads, and classified the detected vehicles into their corresponding make, model, year, and color.- Compared the performance of ResNet50, ResNet101, and ResNet152 on the image classification task in terms of obtained accuracy and inference time