Sami Xiberras

Nlp Engineer @Euler Hermes

Puteaux, FR

Designed and implemented an innovative automated news monitoring system for credit analysts to proactively detect events affecting companies\' credit risk profiles. This full-scale NLP solution processes news articles daily with near real-time analysis.Key Achievements:• Developed specialized Named Entity Recognition (NER) models to accurately identify and disambiguate primary companies mentioned in articles, achieving 80%+ F1 score• Created sentiment classification models for English, French, and German articles, reaching 85% F1 score for sentiment analysis and 90% F1 score for importance classification• Enhanced the system with targeted sentiment analysis capabilities to independently evaluate multiple companies mentioned within the same article• Designed generative AI-powered annotation pipelines for NER, importance and targeted sentiment analysis, creating annotated articles with minimal human intervention• Designed and deployed a complete AWS-based architecture using Lambda functions, S3 storage, data streams, and Elasticsearch for article indexing and retrievalTechnical Environment:• Generative AI: Advanced prompt engineering for synthetic data generation, fine-tuning LLMs for domain-specific tasks• NLP: Fine-tuned BERT-based models, custom NER systems, multilingual sentiment analysis• Cloud: AWS (Lambda, S3, Kinesis, Elasticsearch, SageMaker for real-time inference)• Machine Learning: Supervised classification, transformer models, LSH for semantic similarityInnovation:• Leveraged generative AI to transition from global sentiment analysis to targeted entity-level analysis for complex multi-company articles• Created a novel LLM-powered workflow for generating high-quality financial news annotations specifically tailored to credit risk analysis• Developed a distillation approach combining large-scale LLMs for synthetic data creation with efficient fine-tuned models for production