A few months ago I entered the world’s largest hackathon with over 130,000 competitors.
My original project idea had nothing to do with AI memory preservation.
I started by building a social commerce platform called OAK where people could upload clothing designs and have them printed onto real garments and shipped. Halfway through the competition I realized I wasn’t emotionally connected to the idea, and honestly it didn’t feel differentiated enough to compete at that scale.
One of the hackathon sponsors was Tavus.
Around that time I remembered a voice note my uncle sent me before he was killed in Nigeria.
For years I replayed that recording during difficult periods of my life. Hearing his voice still carried emotional weight long after he was gone. But eventually I kept thinking about something unsettling:
the voice note was static.
I could replay it forever, but I could never ask a new question.
I could never hear a new story.
The memory was frozen in time.
That realization became the foundation for Echovault.
I pivoted completely and spent the next 9 days building an AI digital legacy platform designed to preserve a person’s memories, personality, stories, and voice for future generations.
The Stack
The project started on Bolt.new since the hackathon required it, then eventually moved into Cursor for faster iteration.
The stack ended up looking like this:
- React frontend
- Supabase backend
- Gemini for NLP and embeddings
- Tavus for conversational video
- ElevenLabs for voice synthesis
Mostly written in typescript and deployed on vercel. The core idea behind Echovault was building an “Echo,” an AI representation of a person trained on their memories, personality traits, and life experiences through guided conversations.
The Disaster Moment
At one point during development, an AI assistant accidentally triggered a database reset during a prompt chain.
I blindly approved it.
The entire database disappeared instantly.
That mistake cost nearly two full days of recovery work. I had to rebuild from an old schema.sql file inside the workspace, recreate missing tables manually, and rewrite parts of the RLS policies to secure the database properly again.
The recovered database was never quite the same afterward, but the project survived.
The Result
Echovault ended up winning the “Most Inspirational Story” award at the hackathon.
Part of me secretly hoped to win something more technical because I was proud of how ambitious the architecture became under such a short timeline, but the award still meant a lot because the story behind the product was deeply personal.
The experience completely changed how I think about software creation.
Long term, my goal is to make Echovault a legitimate way for people to digitally preserve themselves for the far future.
Top comments (0)