Search papers, labs, and topics across Lattice.
This paper investigates retrieval challenges in low-resource public service information systems, focusing on food pantry access. The authors built a conversational retrieval system using web scraping, indexing, and a RAG pipeline to answer natural language queries about food pantries. A pilot evaluation with community-sourced queries revealed limitations in retrieval robustness, handling underspecified queries, and grounding over inconsistent knowledge bases.
AI-powered systems for accessing critical public services like food pantries struggle with retrieval robustness and inconsistent data, highlighting the need for more resilient IR techniques.
Public service information systems are often fragmented, inconsistently formatted, and outdated. These characteristics create low-resource retrieval environments that hinder timely access to critical services. We investigate retrieval challenges in such settings through the domain of food pantry access, a socially urgent problem given persistent food insecurity. We develop an AI-powered conversational retrieval system that scrapes and indexes publicly available pantry data and employs a Retrieval-Augmented Generation (RAG) pipeline to support natural language queries via a web interface. We conduct a pilot evaluation study using community-sourced queries to examine system behavior in realistic scenarios. Our analysis reveals key limitations in retrieval robustness, handling underspecified queries, and grounding over inconsistent knowledge bases. This ongoing work exposes fundamental IR challenges in low-resource environments and motivates future research on robust conversational retrieval to improve access to critical public resources.
