ChatGPDune

Dune themed RAG based LLM ChatBot. | Using Ollama, DeepSeek, Neo4J and LangChain

Grupo: Gustavo Henrique, Icaro Botelho, Maruan Biasi, Mauricio Nunes

---

To run:

Ollama Setup:

• Install Ollama from https://ollama.com/download
• Pull whatever model you want to use, by default the project uses only deepseek-r1, you can pull it using the following command:
• pull deepseek-r1:latest

Neo4j Setup:

• Install neo4j desktop from https://neo4j.com/download/
• Open Neo4j desktop and create a database, preferably called chatgpdune
• When creating the database, make sure to add your user and password to the .env

Project Setup

• clone git repo
• create and activate venv via python -m venv venv
• install python requirements via pip install -r requirements.txt

Database Setup

• you need a populated database to use the RAG
• the easiest way is to import the pre-processed embeddings from the Dune 1 book using the csv file
• open neo4j desktop, connect to the database and import the file located inside our project at /database/Ingested/book-1-only/node-export.csv
• This csv file contains all embeddings from the Dune 1 book, which would take hours to process.
• If you want more data inside you database, which is recommended, follow the tutorial below on how to process data locally.

Run Backend

• from project root do the following
• cd backend
• uvicorn app:app --reload

Run Frontend

• from project root do the following
• cd frontend/chatgpdune
• npm i
• npm run start

(Optional) How to process data locally for the RAG

• To process data locally, which is basically generate the embeddings for chunks of text using your own machine, do the following
• Pull the nomic-embed-text model via ollama using ollama pull nomic-embed-text
• cd RAG
• Everything inside the "data" folder will be processed and it's embeddings will the added to the database.
• If you want to ignore a file, which means, skip it's embeddings, you can put it inside /data/ignore. All other children folders inside /data will be processed, only /data/ignore wont.
• It supports the following file formats: [.txt, .pdf, .html, .htm, .md, .markdown]
• It also does webscraping from the dune fandom wiki, you can choose to turn this feature on or off when you run the data ingestion system.
• Inside the file "data_ingestion.py" you can customize different parameters on the "Tweakable settings" section.
• On line 294, you can customize how many wiki pages to scrape.
• To run the data ingestion system, use the following command: python data_ingestion.py