📖 Origins of the Idea

This project began during my studies at Johns Hopkins University, in the course Generative AI for the Humanities and Social Sciences. In Week 14, we explored Multimodal Retrieval-Augmented Generation (RAG) — a framework that combines text and image retrieval with large language models.

Inspired by this, I asked myself:

“Could these methods be applied to archaeology, allowing us to rediscover ancient knowledge in new ways?”

The answer became the Olmec Town Prototype, grounded in my own research: The Olmec Town, a thesis on the origins of urbanisation in Mesoamerica.

⚙️ How the Prototype Works

The system was designed as a research assistant for ethnoarchaeology:

Ingest academic sources (text + images from the thesis).

Embed data in a vector database (text via transformers, images via CLIP).

Retrieve relevant passages and images in response to queries.

Generate natural language answers with GPT-4.

Evaluate results with an “LLM-as-a-Judge” for groundedness and relevance.

➡️ [Insert Workflow Diagram / Images Here]

The interface was built in Gradio, making it possible to interact with the thesis through natural language queries.

🏛️ Recognition and Academic Evaluation

The prototype received both professional review and university evaluation, highlighting:

✅ Strong alignment with advanced RAG techniques taught in the course.
✅ Thoughtful chunking and embedding strategy for dense academic texts.
✅ Integration of cross-modal retrieval (text + image).
✅ Mature evaluation via LLM-as-a-Judge.
✅ A modular, clean workflow — suitable for academic demonstration or publication.

Suggested refinements included adding metadata outputs, commenting out test cells for clarity, and expanding to additional Mesoamerican datasets.

📅 Milestones and Timeline

Spring 2025 → Project developed as part of the Johns Hopkins course.

July 2025 → Prototype received professional feedback and was recognised as research-grade.

September 2025 → Scheduled presentation at the Cultural Heritage Conference, introducing the wider community to this approach.

2025 onward → Evolution into HeritageLens, an expanded system to support broader cultural heritage research.

🔭 Why It Matters

This project demonstrates how Generative AI can serve the humanities:

Making complex archives accessible to researchers and the public.

Enabling cross-modal exploration of text, images, and artifacts.

Opening pathways for comparative and interpretive research across sites and disciplines.

What started as a university assignment has evolved into a research-grade prototype — and the foundation for a broader mission: bridging ancient knowledge and modern AI, ensuring that cultural heritage is studied, preserved, and reimagined with care.

✨ HeritageLens — The Next Step

The Olmec Town prototype was only the beginning. It has since grown into HeritageLens, a vision for multimodal AI systems that:

Support researchers in archaeology, history, and cultural heritage.

Provide intuitive interfaces for querying complex materials.

Encourage collaboration between human interpretation and machine assistance.

➡️ HeritageLens will be showcased at the upcoming Cultural Heritage Conference and continues to evolve as part of the InsideGPT research line.