One Piece AI Chatbot

#

This project demonstrates how I built a Retrieval-Augmented Generation (RAG) chatbot using structured data.

Problem

#

How can we build a chatbot that answers domain-specific questions accurately without hallucinating?

Solution

#

I built a RAG system using:

• Structured Markdown data
• Vector embeddings
• Semantic retrieval
• Large Language Model (LLM)

How it works

#

1. Data was collected and structured into Markdown
2. Content was split into smaller chunks
3. Each chunk was embedded into a vector database
4. User queries are matched with relevant chunks
5. The LLM generates answers based on retrieved context

Data structure

#

The dataset is divided into:

• Characters
• Arcs
• Devil Fruits

Example:

# Character: Monkey D. Luffy

## Role
Captain

## Abilities
- Rubber powers
- Haki

Results

#

The chatbot can answer:

• Character information
• Story arcs
• Abilities and powers

With improved accuracy compared to a standard LLM.

What I learned

#

• Structuring data for AI systems
• Importance of chunking
• Prompt engineering basics
• Retrieval optimization

Improvements

#

• Custom frontend instead of iframe
• More data sources
• Metadata filtering