Wessim Slimi

Designed and deployed an agentic AI framework (LangChain/LangGraph) on top of Kafka + Flink streams for autonomous maritime use cases across 250+ vessels. Implemented embedded ML models in Flink with ONNX and built RAG + LLMOps/agent evaluation, achieving +15% fuel savings and <10ms inference latency.

Built and operated a SmartShip real-time streaming platform and stateful Apache Flink pipelines with exactly-once semantics, RocksDB state, and checkpointing. Designed the Kafka backbone at fleet scale and delivered Iceberg lakehouse sinks plus hybrid Flink/Spark batch-and-stream pipelines, including an AWS cloud migration.

Refactored monolithic SQL batches into Kafka + Flink streaming jobs, supported by Spark batch pipelines and orchestration via AWS Lambda and Step Functions. Bootstrapped the real-time backbone (AWS MSK/Kinesis/Glue), built feature engineering pipelines to Iceberg, and optimized cloud cost.

Developed and deployed predictive maintenance models for wind and solar assets using sensor telemetry, including automatic shutdown for ice detection. Built ingestion and transformation pipelines for ML training features on Azure/Databricks and applied constrained convex optimization using Gurobi and CPLEX.

Built supervised churn models for auto insurance (logistic regression, gradient boosting, and calibration) and integrated them into an internal pricing engine. Developed dynamic pricing and price elasticity analysis with SHAP-based explainability, using end-to-end PySpark pipelines on Databricks with MLflow for tracking and industrialization.

Developed computer vision models end-to-end, including data labeling, TensorFlow/Keras training, and deployment via Flask REST APIs for real-time classification. Performed internal AutoML (hyperparameter tuning/model selection) and integrated data using GCP services such as BigQuery and Firebase.

Engineering degree in Mathematics & Computational Engineering at IMT Atlantique (formerly Mines Telecom Bretagne), covering topics such as machine learning/deep learning, constrained optimization and operations research, and stochastic modeling. Coursework also included object-oriented programming plus SQL/NoSQL databases, with development in Python, Java, and C++.