Echoes of Yesterday：AI 數碼記憶保存平台

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

Echoes of Yesterday is an offline, AI-powered digital memory preservation platform designed to preserve personal stories, voices, and experiences for future generations.

Inspired by the idea that memories should never disappear, I wanted to explore how AI could help preserve the voices and stories of people we love.

Users can record conversations with family members or upload existing audio recordings. The application transcribes these recordings, extracts meaningful information, organizes stories into searchable memories, and creates a personal digital archive.

Users can ask questions such as:

What stories did Grandma share about her childhood?
Which festivals did Grandpa enjoy celebrating?
What memories do we have about our old family home?

Instead of generating fictional responses, the system retrieves relevant information from actual recorded memories.

The goal is simple: People may leave, but their stories don't have to disappear.

Code

GitHub Repository: https://github.com/afreensumai64/Echoes-of-Yesterday

How I Built It

The project uses an entirely local AI pipeline, combining open-weight language models and open-source tools.

Technology Stack:

Language: Python
Speech Recognition: Whisper.cpp
Open-weight LLM: Qwen2.5-1.5B-Instruct
Local Inference: Ollama
Embeddings: Sentence Transformers
Vector Database: ChromaDB
Backend: FastAPI
Frontend: React + TypeScript
Database: SQLite

Working Pipeline:

Audio recordings are uploaded or captured through the application.
Whisper.cpp transcribes the recorded conversations locally.
The language model extracts names, locations, events, and meaningful stories.
Sentence embeddings convert memories into searchable vector representations.
ChromaDB stores the embeddings for semantic retrieval.
Users can search their archive using natural-language questions.
The system retrieves relevant original memories and generates answers grounded in those recordings.

Why Does Open Innovation Matter?

Privacy is particularly important when dealing with personal conversations, family recordings, and memories.

Using open-weight models and local inference makes it possible to process sensitive recordings without depending on external AI APIs.

Open innovation provides several advantages:

Privacy: Personal recordings and transcripts can remain on the user's device.
Offline accessibility: Once models are downloaded, the core AI pipeline can operate without an internet connection.
Model flexibility: Developers can replace or customize models without rebuilding the entire application.
Cost efficiency: Local inference eliminates recurring per-request API charges.
Transparency: The processing pipeline can be inspected, modified, and extended by the open-source community.

A closed API might provide convenient transcription and summarization, but it would introduce external service dependencies when handling deeply personal data.

Open-source AI makes it possible to build a memory-preservation system where users retain control over their own stories.

The Spirit of It

The inspiration behind Echoes of Yesterday comes from a simple thought:

What happens to all the little stories we hear from our grandparents, parents, and loved ones?

The childhood incidents, forgotten traditions, family histories, favourite places, and conversations that make someone who they are.

So much of this disappears simply because nobody thought to preserve it.

I wanted to build something that uses AI not just to generate content, but to protect something meaningful.

A voice can fade. A memory can become distant. But a story, when preserved, can continue to travel through generations.

That is the spirit of Echoes of Yesterday.

Prize Categories

Open-Source AI
Open-Weight Models
Local AI / Privacy-First AI

Built with open-source innovation for Hacktoberfest 2026 – Week 4: Ghost in the Machine.