AI-based mission-critical applications require efficient configuration mechanisms to meet the strict operational demands of resource-constrained edge environments. FM4AIM introduces a novel approach for generating valid configurations for AI-based applications that are decomposed into microservice chains, employing Feature Models (FMs) to capture complex dependencies and workflows. A major challenge with large FMs is the high computational cost of identifying valid configurations. FM4AIM addresses this issue in two innovative ways. First, it employs a two-stage process for identifying executable configurations at the edge: an offline phase that precomputes configuration data and an online phase that generates executable configurations in real time while considering the current resource availability at the respective edge. Second, FM4AIM utilizes a novel slicing technique that decomposes FMs into Partial Feature Models (PFMs), significantly reducing storage requirements and computational overhead. Our evaluations demonstrate that, in the offline phase, FM4AIM identifies valid configurations up to 23× faster—even for small configurations—and simultaneously enables a 100× reduction in storage requirements, depending on FM complexity. FM4AIM speeds up to 1,000× in the online phase compared to existing approaches.