Page 110 - tech fest 2025 ב״ש חוברת תקצירים

110
Efficiency and Improvement of Production in the Production Department of the “Spectronix” Factory
IEM-C-07
Efrat Suissa; efrat19980@gmail.com Noam Abitbul; noamabitbul1@gmail.com
Advisor: Ms. Nofar Raviv
SCE - Shamoon College of Engineering, Be’er-Sheva
This project was conducted in the Production Department of the “Spectronix” Factory in Sderot, which manufactures various types of fire detectors, and focused on our gas-based model. At first, our goal was to identify deficiencies, such as bottlenecks in the production process. Additionally, we measured the time required to produce a single unit and analyzed the duration of each step within the manufacturing process. We developed a forecasting model based on standard error metrics to predict future demand using the historical data provided by the factory. Finally, an optimized inventory policy was defined, relying on the projected demand forecast, while emphasizing ordering costs to enhance inventory management efficiency. We examined process improvements by replacing several manual operations with a robotic system. These changes should reduce the defect rate, shorten processing times and positively impact the overall production workflow.
Keywords: evaluation of improvements, forecasting, optimization of the production process, simulated production processes, suitable inventory policy, work sampling
Real Estate Search System Based on Vector Databases
IEM-C-08
Nevo Betesh; nevobetesh@gmail.com Nitzan Tzioni; nitzazi@ac.sce.ac.il
Advisor: Dr. Dima Alberg
SCE - Shamoon College of Engineering, Be’er-Sheva
Traditional real estate platforms often fail to capture user intent due to rigid filtering. This project offers an alternative approach through vector-based semantic search and AI models. By utilizing vector databases and ‘large language models’ (LLMs), the system transforms free-text property descriptions and images into high-dimensional embeddings, enabling flexible and context-aware recommendations. This approach allows users to discover listings that better reflect their true preferences, even when specific filters are missing. A multimodal analysis enriches the system’s ability to identify key features relevant to target audiences, such as students. Our goal is to develop a scalable, intelligent search experience tailored to unstructured real estate data.
Keywords: embeddings, LLMs, real estate, semantic search, vector databases