Year: 2026
Type: dissertation
Abstract: Deploying LLM-based AI agents in real operational environments is harder than just building them. This thesis documents the design, implementation and deployment to production of an agentic system at Alan, the leading health-tech scale-up in the French ecosystem. We built it to solve the problem of automating one of the long tail operational processes so common to many organisational environments: Blocked Movement resolution. Since a system like this needs perpetual updating and improvement our work revolves around to axis: the agentic architecture that allows us to solve the problem at hand and a peripheral operational platform built to have operations personnel be the main driver in this agent's evolution. This thesis proposes that the limiting factor for agentic systems in production is not the underlying model any more but rather the organisation's capacity to maintain them, evaluate them and make them evolve. The solution described here is an attempt of taking this affirmation seriously.
Author: Recalde Tarron, Martí
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7163728614