This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine
Team Members: @kkyo, @gifernandessilva, angelina
What I Built
MEMORIA is an open-source historical heritage discovery platform centered around São Paulo's historic Consolação Cemetery (founded in 1858)—South America's premier open-air museum of funerary art and architecture.
For the "Ghost in the Machine" theme, I interpreted it both literally and metaphorically: breathing digital life back into the "ghosts" of the past (poets, modernist painters, architects, and revolutionaries like Mário de Andrade and Victor Brecheret) through an intelligent, context-aware RAG assistant called MEMORIA AI. Instead of letting history fade into cold stone, MEMORIA AI acts as the voice in the machine, grounded strictly in primary archival documents, tomb registries, and historical catalogs.
Demo
- Video Walkthrough & Demo: Watch the walkthrough on YouTube
Code
The project is fully open-source, modular, and built with modern web standards across a decoupled architecture. You can explore the repositories directly on GitHub:
📦 Frontend Repository: github.com/ASangelina/memoria-front
⚙️ Backend Repository: github.com/kkyo1/memoria-back
How I Built It
MEMORIA is engineered as a production-ready Single Page Application (SPA) with progressive web capabilities (PWA):
- Frontend Framework: Vue 3 (Composition API) with TypeScript for robust type safety and clean component architecture.
- Styling & Design System: Tailwind CSS, utilizing a sophisticated editorial palette (Parchment #F9F6F0, Deep Moss Green #1B3B2B, and Antique Bronze #C5A059) paired with Playfair Display and Plus Jakarta Sans typography.
- Interactive Mapping: Leaflet.js with custom-styled geographic markers, popups, and direct hooks into the AI assistant.
- Open-Source AI & RAG Pipeline: Built around open-weight models (such as Gemma 4 for local/edge inference). The AI utilizes a strict Retrieval-Augmented Generation (RAG) service that returns verified archival citations (document name, page number, and direct excerpts) alongside contextual entity links. If a user asks something outside the historical record, the model refuses to hallucinate, maintaining strict historiographical integrity.
Why Does Open Innovation Matter?
Open innovation and open-weight models were absolute prerequisites for what MEMORIA achieves. A closed, proprietary API would lock cultural archives behind black-box servers, restrict custom system prompt engineering, and introduce unpredictable risks of cultural hallucination.
Using open-weight models and open-source frameworks allowed me to:
- Enforce Absolute Grounding: Build a deterministic citation layer that forces the AI to cite specific municipal archive pages.
- Data Privacy & Local Control: Ensure historical documents and heritage records remain sovereign and transparent.
- Community Extensibility: Allow historians, developers, and cultural institutions worldwide to fork the repo, plug in local cemetery archives, and adapt the platform to any historical site on Earth.
The Spirit of It
"We are not building a map of the dead. We are building a way for the living to remember."
The magical moment when the project truly "clicked" was during testing on the map explorer. Clicking on Victor Brecheret’s modernist sculpture O Sepultamento or Mário de Andrade’s resting plot immediately opened MEMORIA AI with precise archival citations and excerpts. In that exact second, the machine stopped feeling like cold code—it felt like stepping onto the historic grounds of Consolação and having a dedicated archivist guiding you through the whispers of Brazil's modern art revolution.
Prize Categories
Best Use of Gemma
Top comments (0)