This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine
👻 Haunted Séance: An Offline AI Experience Where the Machine Talks Back
What I Built
Haunted Séance is an interactive, privacy-first horror experience inspired by the idea of a ghost in the machine.
I wanted to build something beyond a conventional horror website—an experience that combines open-weight AI, interactive storytelling, procedural audio, digital remembrance, and physical hardware.
The project brings together four main experiences:
- 👻 Haunted Terminal: A possessed terminal that responds to user questions with eerie, AI-powered conversations and visual effects.
- 🕯️ Virtual Ofrenda & Remembrance Garden: A digital space inspired by Día de los Muertos, designed to honor memories without impersonating deceased individuals.
- 📻 Offline Séance Engine: Local inference using open-weight AI, allowing conversations to happen on-device.
- 📜 Local Oral History Vault: A zero-telemetry remembrance archive using browser
localStorage, with JSON export for preserving memories.
The experience also includes procedural horror audio, heartbeat effects, spectral whispers, jump scares, CRT effects, and an Arduino-powered LED matrix.
Demo
🔗 Live Demo: https://haunted-seance.onrender.com/
Code
💻 GitHub Repository: https://github.com/shettysaikumar20/haunted-seance
How I Built It
I used Python, JavaScript, CSS, and the Web Audio API to build the experience.
The AI component supports Google's open-weight Gemma 2 (gemma2:2b) through local inference using Ollama.
For hardware integration, I used the Arduino UNO Q, combining its Linux-based Qualcomm QRB2210 processor, STM32U585 microcontroller, and built-in 8×13 LED matrix through Bridge RPC.
The LED matrix displays four patterns: skull, eye, cross, and marigold.
Deployment is configured through Render's render.yaml Blueprint, with dynamic port handling and static asset delivery.
Why Does Open Innovation Matter?
This project explores four important ideas:
- Privacy: Personal memories and family stories should remain under the user's control rather than being sent to external AI services.
- Offline accessibility: Local AI inference makes the experience possible without depending on continuous cloud connectivity.
- Digital longevity: Local JSON exports help preserve oral histories and memories beyond the lifetime of a particular application.
- Zero token costs: Open-weight local inference removes per-request API token charges.
Open innovation made it possible to combine AI, browser technologies, local storage, and physical hardware into one creative experience.
My Agent Session
The project involved AI-assisted development and experimentation with open-weight models and hardware integration.
The Spirit of It
A machine doesn't need to imitate someone we have lost to help us preserve their memory.
I wanted to explore the boundary between technology, remembrance, and the unknown—creating an experience where the machine can tell stories, preserve memories, and respond without claiming to bring anyone back.
Sometimes, the most unsettling thing isn't what the machine knows.
It's when the machine starts talking back. 👁️
Prize Categories
- 🥇 Best Use of Render
- 🥇 Best Use of Arduino
- 🥇 Best Use of Gemma
Top comments (0)