Unleashing the power of generative AI in tool migration for enhanced security

Cybersecurity increasingly relies on strong Access Management to protect applications and sensitive data. Modern solutions improve authentication and user experience, yet organizations still face difficult migrations from legacy systems: processes are complex, timelines are long, specialist skills are needed, and heavy human coordination often causes delays and inefficiencies. Generative AI offers a practical way to reduce this friction. A single conversational assistant can centralize many manual steps, provide real-time guidance, and standardize decisions across onboarding and migration. This can cut redundant actions, shorten implementation time, lower the expertise barrier, and create measurable business value. This thesis explores how GenAI can support Access Management, with a focus on migration to modern platforms. It does so by implementing a virtual assistant using LangGraph: a graph of connected nodes, each with a clear role in processing the user’s input. The assistant drives the conversation and answers users’ doubts and questions, while also addressing security concerns by avoiding the leakage of sensitive information through LLM-based guardrails. The implementation motivates two further lines of work: building a suitable dataset that reflects real-world questions to test performance, and studying how to keep the conversational context small (on the order of a few kilobytes). The thesis also examines how behavior changes when using different LLMs.