AFTERWORLD: Every Place Has a Past. Every Place Has a Possible Future.

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

AFTERWORLD is an editorial, open-weight AI web companion that transforms physical landmarks, urban structures, and historical sites into living portals across time.
Most modern web experiences are designed to maximize screen time, trap users in algorithmic feeds, and pull attention inward. AFTERWORLD does the exact opposite: it exists to pull you off your screen and out into the real physical world.

How it gets people out to "Touch Grass":

• Geolocated Discovery ("Find Portals Near Me"): With a single tap, AFTERWORLD uses your device's geolocation to discover historical landmarks around your immediate physical coordinates. Whether you're standing in front of a centuries-old monument or walking down a historic city street, it turns your physical walk into a field expedition.
• Three Portals of Time:
  1. Past (Strictly Grounded): Deep historical narratives constructed exclusively from verified public records (Wikidata, Wikipedia, UNESCO, and curated archives). Every claim includes clickable, transparent source citations.
  2. Present (Architectural & Physical Reality): Real-world coordinates, satellite/cartographic mapping, physical conservation challenges, and verified imagery.
  3. Future (Speculative Projection): Imaginative horizons labeled Imagined future — not a prediction exploring ecological shifts, architectural preservation, and future human stewardship over decades and centuries.
• Hands-Free Audio Walk Narration: Built-in voice synthesis (Web Speech API) turns every portal into an ambient audio walking tour. Pocket your phone, put on your headphones, and listen to the past and future of the ground you are walking on while looking up at the world around you.
• Zero Distraction, Zero Surveillance: No accounts, no likes, no gamified leaderboards, and zero GPS tracking. You explore physical reality on your own terms.

Demo

Live Demo: [https://afterworld-hacktoberfest-open-source-ai.onrender.com/]
Video Demo: [https://youtu.be/hjsaewygV5U]

Code

GitHub Repository: https://github.com/Mritunjaii/AfterWorld-Hacktoberfest-Open-Source-AI-Challenge-Week-1.git

How I Built It

AFTERWORLD is built around Google Gemma 2 (gemma2:2b) running via a swappable Ollama local inference provider:

1. Zero-Hallucination Historical Grounding

Generative AI often hallucinates historical facts, inventing dates, names, and battles. In AFTERWORLD, the Past mode enforces an uncompromising architectural contract:

• The backend queries Wikipedia/Wikidata APIs for the target location to build an evidence bundle of verifiable facts and source URLs.
• The prompt instructs Gemma 2 to act strictly as a curator, synthesizing prose only from the supplied facts and returning exact fact IDs used.
• If the AI model is offline or facts are insufficient, the system falls back to a deterministic, strictly cited prose synthesizer rather than fabricating history. ### 2. Fully Local, Private Edge Inference Through the AIModelProvider interface, AFTERWORLD connects directly to an Ollama instance serving gemma2:2b:
• Fast 2B Parameter Footprint: Runs seamlessly on consumer laptops and edge devices without needing a dedicated server GPU.
• Swappable Model Adapter: The application is model-agnostic and can effortlessly run Llama 3, Mistral, or custom fine-tunes with an environment variable tweak (OLLAMA_MODEL=gemma2:2b).

Why Does Open Innovation Matter?

I wanted AFTERWORLD to be more than another application that sends a prompt to a closed AI API.

Open-weight models make it possible to experiment with the storytelling process, change models, control prompts, and potentially run inference on infrastructure we control.

Using Gemma through Ollama provides a path toward local development and self-hosted inference. This can reduce dependence on a proprietary model provider and offer more control over where data is processed.

Open geographic data and a modular source-retrieval pipeline also make the project easier to extend and adapt.

Most importantly, the model is part of the actual product: it transforms retrieved historical evidence into an engaging narrative and creates a separate imaginative future for each location.

The vision is an exploration tool that encourages people to experience the world directly, with AI serving as a guide rather than the destination.

Prize Categories

1. Best Use of Gemma: If the implemented Gemma integration is used for narrative generation.
2. Best Use of SerpApi: If SerpApi is genuinely integrated into the source-discovery pipeline.
3. Best Use of MongoDB Atlas: If Atlas is used for application data and story caching.
4. Best Use of Render: If the application or backend is deployed on Render.
5. Best Use of Sentry: If Sentry is integrated for application monitoring.