The Interviewer Agent

We made an interviewing agent.

We made it for a client with a business called Hey! Food is Ready who needed to help her own clients build their biographies on the company website. Specifically, these clients are chefs who can be booked for different catering needs through the website. This interviewing agent is a part of the chef onboarding process and it came to fruition because writing your own biography is hard, especially when it needs to be in a specific style that draws customers in. We wanted to work out how to make this less hard. We wanted to work out how to turn it into a conversation, which will then be turned into a little personal story.

Although you are, for all intents and purposes, “having a conversation” with this agent it isn’t your typical chat bot. It is not an LLM assistant that will chat with you endlessly, perhaps without reaching the desired conclusion. It is not a customer service experience with a limited well of information that may or may not give you the correct answer. It is a series of prompts instructed to ask specific questions (and to continue asking them if a useable answer isn’t given) until a penultimate prompt (after all questions have a sufficient answer) will wrap the conversation up, and a final prompt will take all the answers given and produce a short bio written in a specific style. There is no set script for the question and answer part, but the prompts are instructed to guide someone through a step by step interview, extracting as much data as possible from an unpredictable source (a human).

For this specific agent the chat process starts with the agent telling you that it’s going to ask a series of questions in order to help write the story. It asks your name and which cuisines you specialise in. Your answer kickstarts the agent and it will ask another question, this time a more in depth one. For each answer you give, it will take it on board and continue with a follow up. If you didn’t manage to answer one of its questions, the agent will try and ask you again for the same information but in a different way. There are specific details required, which each question is designed to extract. When you get to the final question, the agent returns a personal story written in the third person about your experience and background as a chef.

The alternative for this process would be filling in a form, having static and unresponsive questions. But this isn’t filling in a form, this is efficient data extraction with a specific end goal. Forms don’t ask you politely in a reworded question something you might have missed out in the previous answer. Forms don’t behave like a skilled interviewer, easing out useful information for the personal story of the user.

There is an easy scale for freedom of interpretation as well as specificity with this agent. We can modify this in any of the prompts. It all depends on what step by step process we want to get through. Ultimately when the shape of information is unpredictable or unknown, this agent can act as a guiding process which can behave in the similarly organic way of conversation.

It is helpful to think of other ways that this process can be used. Where else can something like this be deployed? Beyond customer onboarding some of our ideas so far include market research surveys where the agent can adjust questions based on the respondent’s previous answers to gather more nuanced insights. Another idea is personalised content creation which could help users create tailored content like personalised letters, product descriptions, or even custom newsletters based on their inputs. Something else that is extremely time consuming is job applications, an agent could assist applicants by guiding them through the process of building a CV or cover letter, asking specific questions to highlight relevant experiences.

There are a lot of ways we can modify and adapt this agent. This is just our first application of it.