Stage - Ingénieur IA Générative - AI4B - Strasbourg

About the Company

Sopra Steria Next is a consulting firm that combines business expertise with technology, placing among the best in digital transformation. We partner with large enterprises and public organizations to help them improve performance and competitiveness.

Role Summary

Since 2024, you will join our practiceAI 4 Business and work on a key internship focused on the safety and robustness of Large Language Models (LLMs). Your main task is to define a testing methodology for LLMs on a real-world use case, preparing attack scenarios, evaluating startup solutions, and aligning with OWASP LLM recommendations.

Responsibilities

• Map risks and threats for the use case, using OWASP LLM Top 10 and internal references (guardrails, DLP, supply‑chain).
• Design malicious prompt attack scenarios (injection, exfiltration, jailbreak) and create a realistic red‑teaming protocol.
• Benchmark solutions and tools, testing startups and offerings such as prompt‑guard, DLP, LLM‑as‑judge, and Promptfoo for prompt‑testing automation.
• Define quality, security, performance, and cost metrics; set up observability and tracing instrumentation for reproducible evaluation (e.g., MLflow LLM Evaluate / GenAIOps).
• Execute tests, score the model’s resilience, and formalise a reusable methodology that checks system robustness.
• Publish a scientific article on the study and deliver a synthesis for consultants.

Bonus

Evaluate biases in an LLM to identify risks when using non‑European models (e.g., Deepseek).

Qualifications

Student in the 5th year of an engineering school, eager to learn, take on concrete challenges, and grow within a dynamic team.

EEO Statement

We are an inclusive and committed employer that actively combats all forms of discrimination and promotes a respectful work environment. We encourage all candidates, regardless of background, to apply.