One Tap Away Chatbot project is about creating a real-time Retrieval-Augmented Generation-based (RAG) conversation agent, utilizing an existing large language model to create a response about gender-based violence (GBV) support organizations in Metro Vancouver. One of the team members, Jae Eun Kim worked as a victim support worker and still volunteers as one for people exposed to violence. Through this experience, she has recognized that individuals who lack a strong support network, especially young people, struggle to access crucial information for relevant organizations. Without someone in their support network to guide them toward helpful resources during difficult times, they are faced with a vast amount of information available online to navigate, contact, and verify by themselves. One Tap Away Chatbot is a technology solution to quickly direct the individuals exposed to GBV to information about relevant organizations.

Nowadays, large language models (LLMs) in the form of a chatbot still have limitations in generating responses about highly localized resources. As an aid to this, RAG enables LLM to extend its knowledge in addition to its trained data, then retrieve information from the knowledge source contextualized for specific use cases. As this project’s knowledge source, we are utilizing a spreadsheet of Metro Vancouver GBV support service organizations, carefully curated by Jae Eun throughout her work and volunteer experiences.

This project allowed us not only to apply our learning about RAG to real scenarios, but to challenge conventional methods of resource provision. Traditionally, resource provision is done by a human member within support network informing about relevant services. However, we propose that this chatbot can act as a non-human agent embedded in websites providing GBV support, such as those for police-based victim service units. We expect that this chatbot would enhance access to information without requiring human intervention or manual research.

NEXT STEPS
Jae Eun Kim will continue working on this project to make this chatbot available as an open-source widget. Jae Eun also plans to contact the GBV support organization that she worked at, to showcase the chatbot and confirm that the chatbot’s functionality meets their expectations for security, ensuring it is not susceptible to exploitation. Upon consultation, she plans to establish guardrails for its performance to prevent the chatbot from hallucinating, or generating information outside of its verified knowledge source. As the guardrails are put in place, the chatbot would be rigourously tested with the stakeholders–service providers or potentially general public members of Metro Vancouver–to evaluate the chatbot’s effectiveness in providing resources. Specifically, the stakeholders could interact with the chatbot with realistic scenarios or intentionally ask ambiguous, complex questions to see how the chatbot handles queries that are both within and outside of its knowledge scope. Ultimately, the team hopes to garner more awareness to this project and bring technology innovation to traditional methods of resource provision.