Internet Graveyard: Bringing Orkut Back from the Dead with Local Gemma

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

What I Built

Internet Graveyard brings dead websites back from the Wayback Machine as a small, browsable memorial. Orkut was where a whole generation of Indians made their first friends online, and it shut down in 2014. Internet Graveyard gathers its archived pages into one place and keeps a strict line between what's real and what isn't:

• 🟩 RECOVERED: original archived text, with the exact snapshot link and date.
• 🟪 RECONSTRUCTED: notes and an epitaph written by an open-weight Gemma model, always labeled and never mixed into the original text.

The ghost can speak, but it has to show where its words came from. When the model can't answer, the page says "the archive is silent" instead of making something up.

Demo

This version rebuilt 6 Orkut pages from 2006 to 2008 (home, About, Help, Terms, Privacy, Login), each linked to its Wayback snapshot.

Code

github.com/AyushMohite08/internet-graveyard

How I Built It

• Python with requests and BeautifulSoup.
• The Wayback Machine availability API finds the closest snapshot for each URL, and each page is fetched raw (id_), without the Wayback toolbar.
• Pages are cached locally, so reruns work offline.
• Gemma (open-weight) writes the historian notes and the epitaph, using either Gemma 3 27B via Google AI Studio or Gemma 4 via Ollama cloud, and it's told to use ONLY the archived text. In this demo run the model call failed, so the AI layer honestly shows "the archive is silent". The recovered layer is fully real.
• The output is a static HTML site, with a green border for recovered text and a pink dashed border for anything the AI wrote.

It's deliberately small: one site and six pages. Next steps are rebuilding forum threads, a historian chat that cites pages, and a meter showing how much of a site was recovered versus reconstructed.

Why Does Open Innovation Matter?

The internet's memory is held by a few organizations, and most of it is being lost. Archives are often personal: old forums, school communities, a friend's blog. Open-weight models like Gemma let anyone rebuild and explain those archives, even fully locally, without handing private memories to a closed service, and anyone can check how the reconstruction was made.