Meet SJTU Echo, a pioneering voice-activated QA agent created to improve the information seeking process for Shanghai Jiao Tong University’s student body. Funded by Huawei, this avant-garde initiative aims to surmount the constraints of standard QA platforms by integrating sophisticated technologies that emphasize user engagement and ease of use.
SJTU Echo has been crafted to tackle the widespread issues affecting today’s QA systems, such as the absence of interactive interfaces, dependence on text-based dialogue, and the complexity involved in browsing extensive university web resources. Our approach is centered on the careful gathering of data and the judicious choice of models for technologies like Retrieval-Augmented Generation (RAG), Text-to-Speech (TTS), and Auto-Speech-Recognition (ASR).
In pursuit of timely and precise responses, we’ve constructed an effective web scraping tool that is adaptable for a wide range of scraping tasks and is notable for its efficiency. Our model selection philosophy is to find equilibrium between cost, performance, and speed. We select moderately sized open-source models to reduce the financial burden of API usage and the leasing of powerful GPU servers. These models not only ensure quick responses but also maintain a level of quality that is satisfactory.
