Smart Lost & Found System: A Multimodal Vision-Language Approach for Intelligent Item Retrieval

To replace email-based lost & found with an AI-powered multimodal platform for intelligent item retrieval.

Existing email-based systems suffer from low engagement, manual matching, no image support, and poor scalability. The project replaces it.

Intelligent recovery of lost personal belongings on academic campuses.

Large academic campuses experience a persistent and largely unsolved problem: the recovery of lost personal belongings. Existing institutional mechanisms rely on email- based broadcasting, which suffers from low engagement, manual matching, and poor scalability. A comprehensive user study conducted across the campus community confirmed widespread dissatisfaction with the current system, with the substantial majority of respondents reporting that they either ignore or skim lost-and-found notifications entirely, and only a small fraction successfully recovering their lost items through existing channels. Thisreportpresentsthe Smart Lost & Found System, afull-stackintelligentplatform that replaces the email-based workflow with a centralized, AI-powered solution. The core technical contribution is the systematic evaluation and deployment of state-of-the- art vision-language models for multimodal item retrieval. We benchmarked four leading models—Open CLIP, Sig LIP, DINOv 2, and BLIP-2—using a purpose-built evaluation dataset spanning 20 campus-relevant item categories. Sig LIP [15] was selected as the backbone, achieving 97.0% Precision@1 with an inference latency of 89.9 ms per image, representing a superior accuracy-to-speed trade-off compared to alternatives. The system supports both image-based and text-based item search through a shared embedding space, enabling users to search by uploading a photo or by typing a natural- language description. User feedback confirms strong adoption intent, with the large majority of surveyed campus members expressing high interest in using the platform.