How does FRENDi work, technically?

At FRENDi, we’re passionate about bridging the gap between state-of-art research and good user experience. Our goal is to bring advancements in research one step closer to everyday life, in the form of a mobile application. In this blog post, I will go briefly through how we are bringing FRENDi to life, by providing it with memory and the capability to personalize.

Medium article diving deeper in to the memory structure


Giving memory to chatGPT

By default, chatGPT has limited memory. When you have a long enough conversation, it might forget things from the start of the conversation. The actual chat.openai.com might even get slower. To solve this problem, we are generating memories from the conversation periodically, and saving them to a database. This allows the chatbot to adapt to the conversation, such that the user does not have to remind the chatbot always about user preferences or other details. Our chatbot has memory inspired by human memory structure. More recent memories are treated with higher priority, making your chatbot adapt over time with the user.


Future development depends on the user

Our goal is to develop the application based on user feedback. We don’t want to work in a silo without acknowledging what features the user actually wants to see - our goal after all is to bridge the gap between the research and the user needs. For this purpose, we have a feedback chatbot, where users can leave as detailed feedback as wanted. We will go through this feedback qualitatively and quantitatively, making sure that we have a good overall understanding of how the application is perceived by the users.


Different profession, different chatbot

Making the chatGPT behave in a certain way requires prompt engineering. We have done the prompt engineering for the user under the hood, so it’s easier to start the conversation with a chatbot that has a personalized tone and touch.


Flexible application architecture

Our application architecture enables us to flexibly pivot to different use cases. We can hook our backend to different platforms, but still maintain the effective approach for customizeable chatbot.