Analyste Red Teaming des modèles d'embeddings - Risques - La Defense, Ile de France, France
Core
Evaluate the robustness of embedding models used in RAG and agentic systems through red teaming exercises to identify vulnerabilities affecting information retrieval quality and AI-driven decisions.
Role type
Junior IC model risk analyst (embeddings & RAG)
Builds
Methodological frameworks for identifying vulnerabilities in vector search and retrieval mechanisms
Domain
Banking + AI safety (embeddings, RAG, vector databases)
Deliverable
production ML models
Required skills
Machine Learning, Deep Learning, Python, Red Teaming, Vector Search, Embedding Models, Risk Assessment, Documentation
Preferred skills
Literature Review, Metric Definition, Workshop Facilitation, Presentation Skills
Technologies
Open Source Embedding Models, Vector Databases, RAG Frameworks, Python, PowerPoint, Web Apps
Responsibilities
Conduct literature reviews on embedding model vulnerabilities; Build experimental environments with open-source models and RAG frameworks; Design and implement red teaming scenarios; Analyze attack impacts (knowledge poisoning, search manipulation, multilingual attacks, evasion); Define resilience metrics; Evaluate banking-specific risks; Document findings and present to the team.
