The Findhelp Chatbot

Challenge: #3 Navigating Public Services The Medi and e.e.r.s have been contracted by Findhelp.org (an online tool to find free or reduced-cost civic services and resources). The problem Findhelp is experiencing is that their database is extensive, not intuitive, and difficult to determine if the proper resource was provided for the issue at hand. Our chatbot: the front-end language component is run on LlamaIndex, recommendations via our recommendation algorithm (e.e.r.s.), the Findhelp data is sorted/categorized/stored in the Snowflake VECTOR, language detection using OpenAI, and backend translation via Google Translator. The chatbot functions as so: A user types a message, indicating they need a civic service for a specific zip code (such as housing assistance) If the AI assistant detects a language barrier, it will opt to swap languages Other methods of language detection: Language setting on phone or other devices The AI providing a list of languages prior to starting e.e.r.s. recommendation system takes the key information from the chat (zip code, service, categories) and provides 3-5 best results to the user out of the 1000’s of resources available for that area. State civic services and resources stored and categorized in the Snowflake VECTOR Scope of project: Starting with 5 states (N.C., S.C., N.J., Maryland, and greater D.C.) and 8 languages. Training model for Mandarin, develop for all 50 states, speech to speech component.