Eya Souissi

– Marketplace Recommendation System (End-to-End MLOps Pipeline): Architected and deployed a full ML pipeline for product recommendation from data ingestion to final inference. – Built an automated Apache Airflow workflow to extract and load data into ClickHouse using Playwright and Gemini-1.5-Flash for automated product data extraction at scale. – Trained a neural network-based recommendation model leveraging collaborative filtering and popularity-based approaches, orchestrated with ZenML and tracked via MLflow; applied unsupervised clustering to segment influencers by interests, gender, and collaboration history. – Implemented real-time monitoring of all pipeline stages (extraction, training, inference) using Prometheus and Grafana. – Influencer AI Assistant Chatbot (ReachBot): Designed and developed a conversational AI assistant enabling influencers to track their own performance metrics and brand campaign results. – Integrated Dialogflow and Vertex AI to deliver a personalised virtual assistant that streamlines brand collaboration management and handles FAQ and privacy policy queries. – Built on a RAG architecture with MongoDB as vector store, Text-to-SQL for structured data querying, and Gemini Flash 2.0 as the underlying LLM. – Instagram DM Automation Pipeline (ReachFlowDM): Fully automated an outreach workflow to contact influencers at scale via Instagram Direct Messages. – Orchestrated the pipeline using n8n and the APIFY Instagram Auto DMs API. – Extracted and filtered influencer data from Kolsquare based on multi-criteria targeting (interests, engagement, niche); automated personalised message campaigns to drive collaboration sign-ups.

– Interactive Dashboard: Developed a generic multi-source dashboard using Plotly and Flask for real-time monitoring and data management across diverse data sources. – KPI Computation: Integrated OpenAI APIs and Prompt Engineering to automate computation of key business KPIs with high precision. – Predictive Analytics: Leveraged PyCaret across Classification, Regression, Clustering, and Time Series modules to extract actionable insights and forecasts from business data.

– Built an end-to-end NLP pipeline for automatic fake news detection on official survey-related content. – Applied full text preprocessing (tokenisation, stopword removal, lemmatisation) using NLTK and SpaCy, followed by feature extraction via TF-IDF and word embeddings. Benchmarked multiple classification models (Logistic Regression, SVM, Random Forest, LSTM, and fine-tuned BERT/CamemBERT) and selected the best-performing model based on F1-score and accuracy.

– Designed and deployed an end-to-end business performance monitoring solution. Built interactive Power BI dashboards to track revenue and key sales KPIs in real time; automated the scheduled delivery of reports to clients via email using SMTP and Power Automate; integrated and published the dashboards directly on the company's ERP platform for centralised access.