Aymane Ouahbi

MLSecOps Benchmark and Validation Study for Airbus:

• Conducted an end-to-end ML/LLM security market assessment, evaluating over 30 commercial and open-source solutions across the MLSecOps landscape.
• Established contact and direct technical exchanges with solution vendors to assess architecture, security coverage, and deployment constraints.
• Shortlisted 4 solutions from the initial list based on a comparative detailed analysis.
• Assisted in deploying the on-prem version of the shortlisted solutions within an AWS VPC.
• Executed extensive testing campaigns on the model scanning, runtime protection, and red teaming capabilities of each shortlisted solution.
• Delivered a detailed benchmarking report summarizing market maturity, tool capabilities, risks, and actionable recommendations for decision makers in the client organization.
• Skills: ML Security, Adversarial Attacks, OWASP, MITRE, Prompt Injection, AWS, etc.

LLM System Development for Airbus:

• Designed and implemented a RAG chatbot proof of concept, connected to large-scale aircraft maintenance and technical documentation databases, supporting subject-matter experts in complex technical analysis.
• Improved retrieval accuracy by implementing advanced semantic search techniques, including hybrid search and cross-encoder reranking.
• Engineered a Generalized Inverted Index search layer, reducing keyword-based query latency from 20+ seconds to 1–2 seconds.
• Delivered a validated proof of concept successfully approved for industrialization and production deployment.
• Skills: LLMs, RAG, PostgreSQL, AWS Sagemaker, Langchain, Langfuse, etc.

• PeopleNet Meta4: Provision of third-party application maintenance (TMA) services for the Meta4 PeopleNet ERP HR software for various clients.
• Skills: Meta4 PeopleNet, SQL, VBA, JIRA, etc.
• Prompt Studio: Working with CGI Innovation Team on the development of Coding Assistants by leveraging LLMs and RAG Systems.
• Skills: LLMs, vLLM, RAG, FastAPI, ReactJS, Langchain, NLP, OVH, etc.

• Risk Analysis in MDPs: Analyze and develop reinforcement learning strategies tailored to min imize the risk of ruin in Markov Decision Processes, ensuring alignment and safety in autonomous space systems.
• Python Module Implementation: Design, code, and test the developed models in Python, resulting in a 2x better Survival time and rate, and 3x better exploration rate.
• Integration with Open-Source Libraries: Seamlessly integrate the module with leading open source libraries
• Skills: Python, Tensorflow, Pytorch, Deep Learning, Reinforcement Learning, etc.