Matija S.

eWaiter Vision is a proof-of-concept project designed to develop and validate a custom AI solution capable of detecting staff (waiter) calls in restaurants. Using computer vision techniques, the system analyzes camera streams to identify specific hand gestures indicating a customer requires assistance, aiming to improve service speed and efficiency in hospitality environments. The system needed to process video streams and detect calls with minimal latency and maximum throughput to be effective in a live restaurant setting, while reliably distinguishing the specific 'waiter call' hand gesture from other movements and gestures within a busy scene. The system performed positively in tests across different environments (indoor/outdoor), with various camera placements, and even in complex scenes involving multiple people raising hands simultaneously or at different times. The PoC was successfully deployed and tested on an AWS EC2 instance, validating the chosen deployment strategy.

This project included the development of an innovative augmented reality framework that enables real-time virtual try-on of wristwatches, operating entirely on edge devices such as smartphones and web browsers. Unlike existing solutions that rely on server-based processing or physical markers, our system uses advanced computer vision algorithms to detect hand landmarks and render 3D watch models with high precision and minimal latency. By processing all data locally, this solution ensures user privacy while delivering a seamless, realistic experience that helps consumers visualize watches on their wrists before purchase. This technology addresses a significant gap in the online retail space for accessories, promising to reduce return rates and increase conversion for watch and jewelry retailers.

This project included the design, implementation, and evaluation of a Retrieval-Augmented Generation (RAG) system for medical question answering. The proposed system integrates a state-of-the-art large language model (LLM) with a vector database powered by Neo4j for fast and efficient information retrieval. For generating high-quality vector representations of PDF documents, Nomic’s embedding model was used. The architecture leverages retrieval and generation, enabling the LLM to generate context-aware responses based on relevant data. The system was validated on two distinct test datasets: clinical trial reports from open-source PDF documents and OPC industry standards documents. The experimental results demonstrated the potential of RAG-based systems in the various domains, highlighting their ability to provide accurate and context-rich answers.