To ensure accuracy in answers, the FAQs were researched and tracked over a period of time to create the document that was stored in Supabase. This tool was also able to store the metadata of user interactions through Slack. The agent would receive a question, search its database, and then provide an answer. The second agent in this system was set on a periodic trigger to analyze all of the onboarding questions and then create a report to discuss the pain points of new hires and make suggestions to improve the process.

I wanted to make sure that the responses were as highly accurate as possible and maintain human intervention as and when needed. The evaluate response agent is there to act as a guardrail, and if it's not confident the response from the agent is in the database, it escalates to a human to respond. An important part of this onboarding process is letting users opt out of using the AI and go directly to the coach if they wish.

The Agent had a positive effect and was able to bring to light a number of core challenges with the onboarding process, particularly around the terminology and processes that new hires would go through in their first month. Slack's interface was useful; however, it seemed more appropriate to embed this within the onboarding eLearning package rather than use it standalone, as it would otherwise degrade the social interaction of the educational programme. The agent generated over 50 queries in the first month, highlighting challenges with the amount of information provided and serving as the catalyst for changes