My Memory Jar

Screen 1- My Memory Jar
Screen 2- My Memory Jar

🎙️ My Memory Jar - Preserving Voices Before They're Gone Forever

Inspiration

The idea was born from two deeply personal experiences:

My grandmother's lost stories: My grandmother was one of the few female doctors of Independent India (1947 onwards). She was one of the 4 daughters living and out and about to become a doctor when females studying to become docs or even stepping out of the city was a huge deal. The partition struck and she found herself on the other side of the border, with little to no family support. She garnered courage and travelled till she made it. She had a fulfilling life and was a treasure of stories. An anthropologist friend of mine interested in storytelling and vocal history wanted to record her experiences and write a paper on it. We planned multiple times but couldn't get it done for one reason or the other. Now that she is no more we will lose all her stories of courage, travel, of her successful career, a hospital she set up in a small town to help people. These may not last beyond my generation and it makes me sad that it will be lost forever.

My son's bedtime stories: My son who is 10 now has a deal with me; that every alternate night I will tell him an 'original story' and the other day he will read his story book as bed time story ritual. Our stories are impromptu, where our silly golden retriever is the Gold Monk, a version of Thor called the ThorDog and that of course makes him the God of Dogs! He also has a character of 'King of smartness' and his stuffed toys do make an appearance and all come with their powers and superpowers. We decided to write these down and I did try and even published one on amazon- but the effort of turning a 30 min bed time story took nearly 2 hours to write, detail multiplied by 4 times per week. I eventually gave up and we started losing track of these wonderfully intricate stories that we couldn't have left to memory and so many of them, hundreds of them by now have been lost forever. It would have been great to preserve them and maybe one day my grandkids would enjoy our silliness.

These experiences made me realize: We have thousands of photos but almost no recordings of the voices we love most.

How We Built It

The 5-Day Sprint with Bolt.new:

Building with Bolt.new as my AI pair programmer, I maintained 100% control through careful prompt engineering across 15 windows of development:

Days 1-2: Core recording system with emotion-based categorization, real-time transcription
Day 3: Voice search implementation, family sharing, story generation
Day 4: All API integrations, blockchain verification, audio narration
Day 5: Bug fixes, deployment preparation, final UX fix

Tech Stack Integration:

Bolt.new: 100% of development - from ideation to deployment
Supabase: Authentication, database, and audio file storage
OpenAI: Whisper for transcription, GPT-4 for story generation
ElevenLabs: Converting text stories into audio narratives
Algorand: Blockchain verification for authenticity
RevenueCat: Freemium model with 10 free memories
Netlify + Entri: Deployment and custom domain (mymemoryjar.com)

Key Features Built

One-Tap Recording: Select emotion → Record → Auto-transcribe
Voice Search: "Play grandpa's war story" - first app to search memories by voice
AI Story Generation: Select multiple memories → GPT-4 weaves them into narratives
Audio Stories: ElevenLabs narrates the generated stories as a collection
Family Vault: Selective sharing with access controls
Blockchain Verification: Every memory is authenticated (ready for production)

Challenges & Solutions

Challenge 1: Bolt.new Code Reduction

Problem: Bolt would simplify code during updates, removing features (1,700 → 1,400 lines)
Solution: Maintained strict version control and explicit "DO NOT remove" instructions

Challenge 2: Preserving Exact Words

Problem: GPT-4 was creating fictional narratives instead of preserving recordings
Solution: Engineered prompts to quote exact words with minimal connecting context

Challenge 3: Voice Search Complexity

Problem: Building natural language search for audio memories
Solution: Combined Web Speech API with fuzzy matching on transcripts and emotions

Challenge 4: Real-time Audio Processing

Problem: Large audio files causing upload delays
Solution: Implemented streaming uploads to Supabase with progress indicators

What We Learned

AI Development Speed: With proper prompt engineering, Bolt.new enabled solo development at team speed
Voice-First UX: Audio interfaces need different patterns - fewer choices, clearer feedback
Emotional Features: Starting with emotion selection increases engagement and memory value
API Orchestration: Combining multiple AI services (Whisper → GPT-4 → ElevenLabs) creates magical experiences
Preservation vs Creation: Users want to preserve authentic voices, not generate synthetic ones

Impact & Vision

During development, I tested it myself and immediately captured:

My mother's cooking instructions in her voice
My son's explanation of his "King of Smartness" character
Stories my father never wrote down

The potential is massive:

500M families worldwide
$2.5B market opportunity
First-mover in voice preservation

But beyond numbers, it's about ensuring no family loses another voice, another story, another piece of their history.

Why This Wins

Universal Problem: Everyone has lost or will lose voices they wish they'd recorded
Perfect Timing: Built just as AI makes this possible at scale
Emotional Lock-in: Once you save grandma's voice, you'll never cancel
Technical Excellence: 8 integrations working seamlessly
Built in 5 Days: Demonstrates the power of Bolt.new for rapid innovation

My Memory Jar transforms fleeting moments into forever memories - ensuring that unlike my grandmother's stories or my son's bedtime tales, no family's voices will be lost again.

Built with ❤️ and Bolt.new in 5 days for the World's Largest Hackathon