SBIR Phase I: Developing a Smart Virtual Assistant-Enabled Sewer Asset Management Tool

The broader/commercial impact of this Small Business Innovation Research (SBIR) Phase I project is to improve the level of service for the nation’s sewer infrastructure systems, save millions of dollars of annual spending on sewer pipe rehabilitation costs, and promote human health as the result of the reduced sewer overflows. This innovation is critical to the commercialization of the data management tool as the proposed virtual assistant can unburden data users of the required coding skills for database inquiries. The successful commercialization of the data management tool will have a significant impact on the local economy by creating jobs in sales, civil engineering, data science, and data analytics, business administration. The collaboration with academia will cultivate the future STEM workforce needed in the artificial intelligence (AI) sector. In addition, the undergraduate and graduate researchers to be hired on the project will be exposed to entrepreneurship activities, which will cultivate their business thinking and entrepreneurial spirit, leading to more startup company creation in the future. This Small Business Innovation Research (SBIR) Phase I project aims to develop a smart virtual assistant-enabled sewer infrastructure asset data management tool to facilitate the implementation of data-driven asset management practices in the wastewater divisions among municipalities. Several technical hurdles will be overcome by the proposed R&D activities. First, the tool should be able to map natural language queries both in text and voice (with different accents and under noisy environments) formats to the correct SQL syntax. Second, the tool should be able to interpret the natural language and precisely retrieve records from specific table/tables, and data field/fields from a large sewer database. Third, the tool needs to identify the correct analysis to present the results either in tabular or graphical formats based on data types, the number of data records, and users' personal preferences. The innovation of the proposed solution goes beyond traditional voice recognition and natural language processing techniques by designing a continuous learning framework, in which cloud large language models (LLMs) and local transformer-based models are integrated to improve the real-time responsiveness of the virtual assistant. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.