Origins

Inspiration

With the Baby Boomers becoming the largest group of elders in history, demand for eldercare in the upcoming years will skyrocket to unprecedented levels. Conditions like Alzheimer’s and dementia are especially devastating as they not only rob individuals of memory and identity, but they also place an immense strain on caregivers and loved ones.

Origins aims to help this issue. Reminiscence therapy has been shown to spark joy, reduce anxiety, and improve overall well-being in people living with memory loss by reconnecting them with meaningful stories, photos, and life events. Improved happiness and reduced negative emotion has been proven to slow the effects of harmful health conditions.

What it does

Our site allows family's of users connect with one another and share their stories. It gives a clear and concise way to assign familial relationships and consolidate any family history to be passed down through generations. Every member has the opportunity to share updates, highlights, and fond memories. This history is then integrated into our RAG pipeline to allow users to chat and learn about their family lineage at anytime. Any user has the ability to ask questions or learn more about specific events.

Additionally, many immigrant families struggle to cross a language barrier and unfortunately, meaningful family history can fall through the cracks. Origins preserves all of this history and gives family members the opportunity to be known by their loved ones in a non-threatening way.

Users can see across 5 tabs. A family feed, that allows members of a family post pictures and share stories. The reminisce tab that prompts users to recall previous information to reinforce the mental connection. Our chat bot for users to communicate with. Family tree and family members to allow user to view their lineage and ancestors.

How we built it

Origins is built with Typescript, React, NextJS, and PostgresSQL. We used Supabase for our backend and auth and Typescript for our frontend.

For our RAG pipeline, all posts, memories, and other context is fed into a central vector database in Supabase and embedded with OpenAI's 'embedding-3-small' 1536 dimension vector embedding via the API. From there whenever the user asks a question to our custom chatbot, we query the embeddings, return the most relevent context, and feed it into a Gemini 2.5 Flash query to return a response to the user. Additionally, we use Gemini 2.5 Flash to generate memory prompts.

Challenges we ran into

It was very difficult to simplify the assignment of relationships while also providing a robust family lineage. In the end we found it best to make the user side of this problem as simple as possible and only allow them to assign their parents. We would then use inference to generate all other relationships so they could then be updated and displayed in the family tree. This gave us the best of all desired options. We also had many ideas that were of general use to a family but strayed away from the theme. In the end I think we picked a good balance of functional familial elements and health based features.

Accomplishments that we're proud of

We are very proud of the AI model we used. It can effortlessly accumulate the information passed throughout our project and redistribute to users in a variety of manners. Each user can get a personalized chat of past experiences that should hopefully mimic a family conversation and deepen the bond between members. We also take pride in the family tree and inference based relationships. Finding the correct relationships was quite difficult in many edge cases, but in the end we have an implementation that takes into account parental relationships and pronouns to give the desired result.

What we learned

We learned how to build a full stack web app and use RAG.

What's next for Origins

Currently some features we wanted to implement had to be left out. We would like to overhaul the family feed section and allow chats between family members to occur through our project. Having a user to user chat feature that also utilizes our RAG AI could alter how families discuss their history is a secondary goal of our project. We would also like to redesign the user profiles to have more detailed descriptions and allow users to see the memories or other information.

Built With

• gemini
• nextjs
• openaiapi
• postgresql
• react
• supabase