View Proposal

---

Proposer

John See

Title

Asking the Muse: Agentic Retrieval-Augmented Generation for Cultural Artifacts

Goal

The aim of this project is to develop an agentic RAG system (prototype) for querying cultural artifacts from museums and/or art galleries. The system should be evaluated quantitatively and qualitatively, with comparisons made against conventional single-step RAG systems.

Description

This project will develop an agentic Retrieval-Augmented Generation (or RAG in short) system that can autonomously plan and execute retrieval tasks when answering questions about cultural artifacts. The system will combine an LLM agent with retrieval tools and curated cultural collections, enabling it to identify relevant sources, refine searches (through reasoning), cross-reference information and generate evidence-grounded responses. There is a plan to work with a local museum in Malaysia (i.e. Penang State Museum) but this has not been confirmed. Fallbacks are readily available - there are other open access data from various museums around the world that can be utilised for this project. The project would also be using as many open-souce tools/libraries/LLMs as possible.

Resources

Open-source tools/libraries/LLMs, high performance compute (not necessary if there are no additional model adaptations required)

Background

Cultural artifacts such as artworks, photographs, manuscripts and historical objects contain rich information that is often distributed across museums, archives and digital collections, in various modalities ranging from documents and images to audio. Recent advances in Retrieval-Augmented Generation (RAG) allows LLMs to retrieve relevant information from external knowledge bases before generating an answer. However, traditional RAG systems typically performs a relatively straightforward search-and-retrieve process and may struggle with complex questions requiring multiple searches, source comparison and multi-step reasoning. An agentic RAG approach can address this by allowing an AI agent to decide what information to search for, which sources to consult, and how to refine its retrieval strategy. This project therefore explores how agentic AI can support more intelligent evidence-grounded exploration and interpretation of cultural heritage knowledge.

Url

Difficulty Level

Moderate

Ethical Approval

Full

Number Of Students

1

Supervisor

John See

Keywords

agentic ai, rag, cultural artifacts

Degrees

Bachelor of Science in Computing Science