Remi | Devpost

Inspiration

When you think of the word “nostalgia”, what’s the first thing you think of? Maybe it’s the first time you tried ice cream, or the last concert you’ve been to. It could even be the first time you’ve left the country. Although these seem vastly different and constitute unique experiences, all of these events tie to one key component: memory. Can you imagine losing access to not only your memory, but your experiences and personal history? Currently, more than 55 million people world-wide suffer from dementia with 10 million new cases every year. Many therapies have been devised to combat dementia, such as reminiscence therapy, which uses various stimuli to help patients recall distant memories. Inspired by this, we created Remi, a tool to help dementia victims remember what’s important to them.

What it does

We give users the option to sign up as a dementia patient, or as an individual signing up on behalf of a patient. Contributors to a patient's profile (friends and family) can then add to the profile descriptions of any memory they have relating to the patient. After this, we use Cohere API to piece together a personalized narration of the memory; this helps patients remember all of their most heartening past memories. We then use an old-fashioned styled answering machine, created with Arduino, and with the click of a button patients can listen to their past memories read to them by a loved one.

How we built it

Our back-end was created using Flask, where we utilized Kintone API to store and retrieve data from our Kintone database. It also processes the prompts from the front-end using Cohere API in order to generate our personalized memory message; as well, it takes in button inputs from an Arduino UNO to play the text-to-speech. Lastly, our front-end was built using ReactJS and TailwindCSS for seamless user interaction.

Challenges we ran into

As it was our first time setting up a Flask back-end and using Kintone, we ran into a couple of issues. With our back-end, we had trouble setting up our endpoints to successfully POST and GET data from our database. We also ran into troubles setting up the API token with Kintone to integrate it into our back-end. There were also several hardware limitations with the Arduino UNO, which was probably not the most suitable board for this project. The inability to receive wire-less data or handle large audio files were drawbacks, but we found workarounds that would allow us to demo properly, such-as using USB communication.

Accomplishments that we're proud of

Utilizing frontend, backend, hardware, and machine learning libraries in our project
Delegating tasks and working as a team to put our project together
Learning a lot of new tech and going to many workshops!

What we learned

We learned a lot about setting up databases and web frameworks!

What's next for Remi

Since there were many drawbacks with the hardware, future developments in the project would most likely switch to a board with higher bandwidth and wifi capabilities, such as a Raspberry Pi. We also wanted to add in a feature where contributors to a patient's memory library could record their voices, and our Text-to-Speech AI would mimic them when narrating a story. Reminiscence therapy involves more than narrating memories, so we wanted to add ore sensory features, such as visuals, to invoke memories and nostalgia for the patient as effectively as possible. For example, contributors could submit a picture of their memories along with their description, which could be displayed on an LCD attached to the answering machine design. On the business-end, we hope to collaborate with health professionals to see how we can further help dementia victims.