ML/LLM Engineer

Architected a production text-to-SQL assistant empowering 40+ financial analysts to interrogate complex datasets without SQL expertise.

• Semantic parsing and query optimisation lifted accuracy to 85.6% with automated fallback logic.
• Decision-making cycles shrank by 27% through faster insights and guardrailed analytics.
• Delivered compliant AWS pipelines spanning Lambda, Bedrock, and OpenSearch with production monitoring.

Brought LLM reasoning to vehicle telemetry, diagnosing anomalies for connected cars within the R&D division.

• LLM-powered diagnostics translated sensor streams into natural-language repair guidance with 72.8% detection accuracy.
• Knowledge-retrieval pipelines mapped anomalies to procedures, cutting misdiagnosis 14.2% and saving €1.1M annually.
• Root-cause analytics reduced technician investigation time by 24.5% across pilot fleets.

Partnered with the TER division to curb multimillion-euro ticket fraud through predictive controls.

• Fraud rate dropped from 7.2% to 5.9% via targeted interventions powered by hypothesis testing.
• Advanced models (RF, XGBoost, LSTM) achieved AUC-ROC above 0.87 for risk scoring.
• Regression forecasts with 14.7% RMSE improved inspector deployment across the rail network.

Advanced operational safety initiatives for France's gas transport network within a 6-person innovation program.

• Forecasting pipelines improved incident anticipation through bespoke gas propagation models.
• Confluence-integrated LLM assistant shortened support response times by 38.5%.
• Established TDD-driven monitoring and retraining practices for ML systems in production.

Scaled ML products for thousands of legal professionals within a dedicated MLOps squad.

• Implemented ML platforms with GitLab CI/CD, ZenML, and MLflow reducing deployment cycles by ~41%.
• Refactored codebases to eliminate tech debt and improve maintainability of legal analytics tooling.
• Managed cloud infrastructure with Terraform and SageMaker for reliable releases.

Strengthened fraud prevention and AML/CTF compliance inside the Risk & Compliance department.

• Implemented AML/CTF detection pipelines that reduced manual workload by 24%.
• Engineered PySpark/Hadoop workflows to elevate data quality across millions of transactions.
• Leveraged Neo4j graph analytics to surface fraud networks and suspicious patterns.

Automated document analysis for innovation labs by combining OCR and NLP workflows.

• Reduced manual processing time by 34% with ML/DL models tuned for printed and handwritten text.
• Improved F1-score by 11.8% through rigorous hyperparameter tuning.
• Boosted classification and deduplication with targeted text similarity analysis.

Supported modernization of citizen-facing digital services with NLP automation.

• Built NLP models for classification, topic modelling, and chatbots serving public administration teams.
• Automated information extraction pipelines improving processing efficiency.
• Developed Python web apps and scraping pipelines for rapid data collection.