Zeineb GHRIB

Context & Tasks: - Contextual Topic Extraction: ◦ Utilized Generative AI alongside traditional ML techniques to extract meaningful topics from conversation transcripts. ◦ Integrated nuanced sentiment analysis (e.g., "Concerned," "Frustrated") for more detailed meeting analytics. ◦ Applied methods like Retrieval-Augmented Generation (RAG) to refine extraction results, providing sales teams with deeper insights. - Small Talk Analysis for Enhanced Sales Engagement: ◦ Built an AI-driven tool to automatically identify small talk topics like sports, holidays, and hobbies from customer conversations, enabling personalized interactions that foster rapport. ◦ Employed a mix of generative AI and traditional ML algorithms, all running on in-house infrastructure to ensure data security and scalability. - Tonal Emotion Detection: ◦ Streamlined and maintained the continuous training cycle of a CNN-based tonal spectrogram model for emotion detection. ◦ Led data quality initiatives, including inter-rater agreement monitoring and error analysis, which significantly improved annotation consistency and model accuracy. ◦ Managed annotated data analysis, continuously optimizing data collection processes to sustain high model performance. - Vector Search Project Supervision: ◦ Guided an internship project that improved the performance of a keyphrase-based search service using vector search technology. ◦ Achieved significant performance gains, reducing execution times by 50%, demonstrating leadership in mentoring and technical problem-solving.

Lead on core conversational AI use-cases: - Improved performance on Speech Emotion Detection models (using calibration techniques, balanced generators, auto-encoders). - Disrupting Topic Modeling approach for survey feedbacks (contextual capturing embeddings, customized distance metrics). - Team mentoring on Lies Detection & Diarization Use-cases. - End-to-end automated Machine Learning Pipelines design and implementation (fasten*10 the ability to iterate) - Cloud-based ML Platform setup (AWS & Databricks)

Results: - Research article accepted by IJCNN conference and published by IEEE: https://ieeexplore.ieee.org/abstract/document/9207013 Context & Tasks: Realize a POC for anomaly detection in time series data: - Design an hybrid based approch for anomaly detection in time series using LSTM autoencoder for seq2seq representation + another classifier such as LinearSVM, SGD... - Develop the full process (from collecting to visualisation) with agile software development (respect PeP8 convention and other programming best practices) using python (and related libraries tensorflow, keras, scikit-learn, seaborn..) - Compare the proposed approch with the state of the art techniques (isolationForest, OneClassSVM, LOF..) - Write a research article that was accepted by IJCNN conference published by IEEE Technical Stack : RNN, LSTM, research writing skills, TSNE, outlier detection, isolation forest, LOF, keras, scikit-learn, seaborn

Context & Tasks: Develop a sentiment analysis unit using machine learning algorithms and integrate it to a full algorithmic trading process - Sentiment analysis of "news" data using text mining techniques, NLP, WordEmbedding Glove) and an LSTM neural network. - Comparing the proposed approch with dictionnary based classifiers Technical Stack : NLP, WordEmbedding Glove, Text Mining, Python, Keras.

Context & Tasks: Design and implement a network intrusion detection system using machine learning algorithms - Evaluate the network security by conducting simulated attacks ICMP Hijacking, IP Spoofing, ..) - Develop the intrusion detection system POC with Python using KDD Cup 99 well known dataset

Context & Tasks: Oracle database administration and supervision - Update existant database, add new users (new employees) - Ensure database access, consistency and integrity - Write database documentation, including data standards, procedures and definitions - Implement different bda tasks such as creating indexes, implementing triggers and Locks with PL/SQL..

Implement a VPN network: - Implement and configure a virtual private network between 2 distant networks using OpenSSL et StrongSwan.