Mohamed Amine Taif

As a data scientist at Publicis Group - World's largest communications group : - Developed a granular statistical Media Mix modeling model based on a bio-inspired optimization algorithm (Ant colony optimization) with access to limited modeling data (Strictly Media and CRM data). - Contributed to analyzing the marketing performance of a client (Maison du Monde) to optimize marketing investment allocation.

Dredge Mining Optimization Project : I worked on testing metaheuristic algorithms to find the optimal mining path taking into consideration potential exploitation diminishing returns.

Worked on two main projects related to NLP and Text classification in the context of flight operations at Air France : - Worked on an unsupervised hierarchical classification LLM based model for pilot reports in the context of flight operations performance, dealing with an extreme number of classes (192 classes) and very limited annotated data. I Implemented a tree of thought based method that optimizes the class names for text classification by generating few class names examples based on the content of the annotated data, evaluating each of the generated class names and critiquing them to branch out from more representative class name generations. - Implemented an LLM based message filtering model that classifies flight operations messages in real time as either information or alert for the TA360 app (Turn Around 360). The predictions consider not only the content of the messages but also the temporal context, such as the time the message is sent and the typical time reference of the activity associated with the message.

worked on the implementation of an open source based RAG architecture for private documents use cases, I was in charge of : - Reviewing state of the art open source embedding models, benchmarks on MTEB retrieval datasets. - integration of the embedding model with Deeplake (open source vector store) in the RAG architecture. - testing the retrieval solution on llama 2 LLM for question answering tasks on private documents.

First internship experience : worked on an assistant chatbot project for the national sector accreditation plateform : - automatization of the plateform's support system with a rule based chatbot. - Pre-processing of a dataset of email exchanges. - Classification of the emails by problem with supervised machine learning models.