Anass Smali

Design and deployment of a multi-label NLP classification system in production for the M2M Malima chatbot, improving the understanding of customer needs on multilingual data (FR, EN, ES, RO).

• Development of a semantic representation pipeline based on OpenAI text-embedding-3-small, combined with supervised models (SVM, MLP, Voting Ensemble) for automatic user intent classification.

• Industrialization of an end-to-end MLOps architecture on Google Cloud Platform integrating Vertex AI, BigQuery, Firestore, Cloud Functions, Docker, Terraform, and GitLab CI/CD, with automated deployment and model versioning.

• Orchestration of an automated daily processing pipeline via Cloud Scheduler, ensuring the execution of processing and monitoring workflows in production.

• Deployment of an ML observability system including performance monitoring (latency, prediction distribution, output quality) and data drift detection to identify changes in user behavior and usage.

• Centralization of predictions and metadata in BigQuery, with the design of real-time dashboards on Looker enabling Business and Marketing teams to track changes in customer needs.

Results: 50% reduction in documentation search time and significant improvement in response quality.