Industry: Internal HR, HealthcareRole: Conversation Designer
Skills: Conversation mapping, system prompt writing, troubleshooting
A quick note: almost all of the projects I work on are under an NDA and are authenticated experiences, so I can’t share particulars here when it comes to names and detailed designs.
The Projects
The Internal HR FAQ Bot was the first implementation of a Generative AI bot for a product company. Because the client wanted this bot to be used by all of its employees, we decided that Slack was the best channel for the bot. The company had a lot of informative, useful information for employees, but it was decentralized and not always easy to find, especially across all regions. The vision for the Slack bot was to bring all of that content to one place and let employees ask questions the way they want, and get the answer they need in a channel they already use all the time.
The main goal for the Internal IT FAQ Bot (for a large hospital in the US) was to provide easy access to FAQ for researchers employed by the hospital and cut down the number of IT tickets submitted. I worked closely with the client to help optimize their content, create a knowledge base, create the system prompt, and test and iterate on the prompt to ensure the LLM was providing answers the way the client needed them to. The chatbot is web-only and currently lives on their intranet.
The Challenge
For both projects, the challenge was similar and completely new to the field. I had to figure out how to incorporate both a “traditional” designed flow (rule-based) with an LLM-powered knowledge base. It’s difficult to pass the flow back and forth between the rule-based flow and the FAQ that the LLM was providing. This needed a lot of iteration, trial and error, and close collaboration with the Developers and Prompt Engineers. We were able to create more elegant handling and create a seamless experience for the user by using a number of backend checks to essentially allow the chatbot to know what the LLM was doing.
The Learnings
These projects also gave me experience with writing and testing prompts. I borrowed a template from a Conversation Designer on LinkedIn and adapted it to suit the needs of the project, which was the first template we had for prompt testing.
