Last Broadcast：以 AI 為博帕爾歷史記憶發聲

Hacktoberfest Open-Source AI Challenge: Ghost in the Machine Submission 👻🌼

Last Broadcast — Giving History a Voice with AI

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

What I Built

Last Broadcast — Bhopal: Voices Across Time is an AI-powered historical archive designed to preserve, explore, and bring local stories to life.

The idea came from a simple question:

What happens to a city's memories when the people who carry them are no longer around to tell their stories?

History is often preserved through books, documents, and photographs, but personal memories and oral histories can disappear over time.

Last Broadcast explores how AI can help preserve these fragments of collective memory.

The platform aims to let users:

Explore curated historical collections about Bhopal.
Upload audio recordings and convert speech into searchable text.
Extract meaningful information from historical narratives.
Ask questions about archived material using AI-powered retrieval.
Listen to AI-generated narration of historical content.

I interpreted Ghost in the Machine as the idea of giving a voice to memories that might otherwise remain buried in documents, recordings, and forgotten stories.

The goal isn't to replace human memories with AI-generated history. It's to build a tool that helps preserve, organize, and make those memories more accessible.

Demo

Live Application: Last Broadcast — Add your deployed frontend URL
Backend API: https://last-broadcast-api.onrender.com
API Documentation: https://last-broadcast-api.onrender.com/docs
Health Check: https://last-broadcast-api.onrender.com/health

Code

GitHub Repository: Last Broadcast — Source Code

The project is built with a modular architecture, separating the frontend, backend, AI services, and data layer.

How I Built It

I wanted to build something that combined open-source AI with a practical application rather than simply wrapping an LLM in a chatbot interface.

Tech stack:

Frontend: Next.js, TypeScript, Tailwind CSS and shadcn/ui.
Backend: Python, FastAPI, SQLAlchemy and Alembic.
Database: PostgreSQL with pgvector for vector-based retrieval.
LLM: Qwen 2.5 7B Instruct through Hugging Face Inference Providers.
Embeddings: sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2.
Speech-to-Text: ElevenLabs Scribe.
Text-to-Speech: ElevenLabs multilingual TTS.
Deployment: Render.

The intended AI workflow is:

Audio recordings are submitted to the platform.
Speech-to-text converts recordings into text.
The resulting content is organized and processed into searchable information.
Embeddings enable semantic retrieval from the archive.
Qwen generates responses grounded in retrieved historical material.
ElevenLabs provides audio narration for an accessible listening experience.

I also wanted the architecture to remain modular, so individual AI components can be replaced without rebuilding the entire application.

An important design principle is distinguishing verified historical information, personal testimony, editorial interpretation