Yahya Ghrab

– Optimized the existing promotion quality model. – Challenged current model design with popular regression models. – Enhanced the Promo Optimizer's capabilities through major code refactoring and performance optimization. – Extended the project scope from PGC to PFT and NONAL segments : SQL + Airflow + python. – Continuously updated the Airflow ETL dags with new tasks. – Worked with DevOps to migrate heavy-processing dags to GCP Batch. – Continuously updated DBT SQL models. – Updated the deployment process (Jenkins to gitlab CI). – Migrated images from JFROG to Artifact Registry and improved images versioning. – Upgraded the whole code-base from python 3.6 to python 3.10. – Upgraded the API interface/code (FastAPI) to provide AI-driven smart suggestions to users.

– Designed and deployed two production-grade LLM-based chatbots, leveraging generative AI to improve customer interactions. – Augmented the chatbots architecture with GPT-based RAG through automatic scraping FAQs and PDF docs. – Benchmarked open-source vector databases: FAISS, Chroma, PostgreSQL(pgvector). – Developed an internal data platform to streamline data collection, transformation and deployment. – Implemented robust ETL pipeline to collect and post-process (annotate) chatbot usage data (gitlab cron jobs - Bigquery - stitch - python). – Built and maintained python scripts to manage the whole data pipelines (data ETL pipeline, testing, training, usage data collection and annotation, model update, monitoring, CI/CD/CT deployment). – Implemented human-in-the-loop annotation using Doccano. – Implemented MLOps practices for seamless integration, training, and deployment of ML models. – Created and maintained a KPI dashboard to monitor user satisfaction and collect new training data. – Continuously monitored chatbot performance through Grafana to track resource utilization (with alerts via Slack, email notifications, and dashboards), complemented by batch-fed usage statistics displayed on a GCP Looker KPI dashboard (overall satisfaction, engagement rate, sessions volume, hot topics, sentiments analysis, fallbacks, etc.). – Led research and development efforts on voice-bot integration using state-of-the-art AI models.

– Collected and analyzed data from legacy tools and SQL databases to optimize automotive supply chain planning. – Developed scenario generation tools and visualized optimal solutions using Tableau and Spotfire. – Applied supervised learning techniques (XGBoost) to enhance supply chain planning models.