Inspiration

When my friend’s mom mentioned the stories that her immigrant Vietnamese mom was aging and her story risked being erased forever, the immediate suggestion that came up was a biographer service that could write her story for $10,000-$15,000 but that too had massive downsides. For starters, my friend’s grandma wasn’t comfortable opening up about vulnerable parts of her past to someone foreign, and as a busy person, not being able to navigate the process at her own pace and not being able to squeeze a quick interview during a 30 minute opening in her day were dealbreakers. Ever since, the idea of an AI biographer has been linkering in my head. Conventional AI biographers that are merely GPT wrappers have fundamental flaws in their text-text offering and in their qualities. They attempt solving a problem but create 20 others in the process like having to complete your whole biography interview in 1 sitting and having limited information in your biography because of the slow nature of typing and reading.

What it does

So I built the first conversational AI biographer system that prioritizes quality and ease of use. You can just call up a phone number and the system can retrieve your completed content, determine the next logical topics to discuss and ask questions to navigate complex and deep topics. With dynamic context injection, it can ‘remember’ important details that are relevant to topics being discussed in the current session. After conducting several sessions and capturing stories through predetermined chronological periods, the Biographer then compiles a biography.

How we built it

To build it efficiently, I broke the larger system down into smaller micropieces that I could stitch together. My main priorities were the database MCP system that handles all data going in and out of the database; the fine tuned Biographer agent that handled voice input to voice output and effectively brainstormed pathways to navigate conversations; the call handling system; hosting platforms; and the connection tunnels as websockets between all of these.

Challenges we ran into

One major challenge I ran into was setting up the websockets between the individual components because the websocket compatibility rulesets for each of these pieces differed. This was a particularly challenging problem with the hosting because webrtc packages needed very specific https rules to be met. I was able to solve this problem with some technical expertise (and luck) through using cloudflare as an intermediary to meet the necessary rules.

Accomplishments that we're proud of

One trait that I showed through this process that I’m proud of is that I was looking at the problem through a scope of scalability and entrepreneurial positioning to set the product up for future success. Instead of just looking for the easiest technical solution to the challenges I faced along the way, I looked at what architecture made the most sense to set up with scalability in mind. I am also particularly proud of the technical rigor of the final product.

What we learned

My most significant learning from this is the value of empathizing with the end user to be able to architect everything from the feature-set to the tech stack to the medium of use. I ended up doing this by picturing myself as an elderly person to come up with the most desirable features (the ability to repeat itself and understand quick language switches or 2 languages in the same sentence); Chaining scalable tools together and making all the parts of the stack that users see easy to understand; and setting up the interviews through phone calls which as accessible and easy to use.

What's next for Biographer

I believe that the biography is primed for full scale deployment and has immense value that it can provide to elderly people. I would like to see this through to commercialization and deployment to have as much impact among elderly people as possible. I would also use the award money (if I win) towards better positioning of the biographer and on advertising to begin getting traction.

Built With

Share this project:

Updates