Achik Aznag

Context: Centralizing and securing genomic data flows (IP) within the Germplasm Security team. - Designed and managed a centralized Azure Data Lake for high-volume genomic data, queried via optimized SQL. - Optimized ETL pipelines (Polars, Databricks), achieving a 70% reduction in processing latency. - Automated data integrity validation and environment deployments using CI/CD pipelines to ensure production reliability. - Modeled data using star schema and built Power BI dashboards for data quality monitoring. - Authored technical documentation and actively participated in code reviews.

Context: R&D for automotive performance optimization via Digital Twins and predictive maintenance. - Engineered real-time "Data-to-Viz" pipelines connecting telemetry data to Unreal Engine and Blender. - Modeled and optimized high-frequency time-series storage in InfluxDB to handle massive sensor data streams. - Implemented Physics-Informed Neural Networks (PINNs) for system degradation forecasting. - Visualized real-time performance metrics using Grafana dashboards.

Context: Digital transformation of mining operations to optimize industrial steering. - Designed distributed real-time data architectures using Apache Kafka and Spark. - Transformed raw IoT sensor data into refined features for process optimization models. - Developed a RAG (LLM) assistant via FastAPI and built a full-stack supervision dashboard. - Integrated Machine Learning models to predict production volumes based on sensor input.

Built and deployed a production French NLP system for customer feedback analytics, based on my own CamemBERT models (88k+ downloads on HuggingFace). -> Designed a self-improving data pipeline — raw feedback in, automated annotation, human validation, and back into the next training cycle -> Took my open-source model from HuggingFace to production on Azure, fully containerized with Docker and orchestrated on Kubernetes -> Worked closely with the marketing team to translate business needs into model improvements, owning the full cycle from data to deployment