DnD Generator

This is a simple project for generating homebrewed monsters and other stuff to be used as inspiration.

I want to use this.

If you are not interested in writing code for this, you can use the application by starting it using Docker and providing an Open-Ai-Key. Follow this:

1. Install Docker
2. Get yourself an Open-Ai-Key from OpenAI. This will need a credit card, but since the Text-Generation used Gpt4-mini, 20 prompts cost like a cent or so, so don't worry. Image-Generation uses Dall-E-3, which costs a little, but you're still ar about 3 cents for an image.
3. Checkout (or copy) the main docker-compose.yml (the one saying # I'm the main one at the beginning)
4. Go into the folder with the docker-compose.yml using some kind of shell (cmd, bash, etc) and run set OPEN_AI_KEY=XXX (Windows) or export OPEN_AI_KEY=XXX (Mac) where XXX is your Open-Ai-Key.
5. Run docker-compose up -d. This starts the backend, frontend and every needed 3rd party service. When this is up, you can now reach the application in a browser under http://localhost:4200.

I want to develop this

You can develop the application locally, but you will still need OpenAI-Api-Key. Go to OpenAI, create a new key and export the value to the OPEN_AI_KEY environment variable. Having that, you can start the spring-boot backend from your IDE. You can start the frontend from the /frontend folder by running ng serve. But since creatures and stuff get persisted to a Neo4j database, you'll have to have one running. You'll also need a Min.io bucket, which you also have to provide, since images get persisted there. You can start both by running docker-compose up in the scripts folder. This docker-compose file is a reduced version of the main one, not having the front- and backend.

Key Features

Mostly I build this application to test working with Spring-AI, so the key features include:

• Interacting with a LLM for text- and image-generation
• Obtaining structured output.
• Using a Vector-DB for RAG (Retrival-Augmented Generation).

Interacting with an LLM

For interacting with an LLM, I use Spring-AI with a GPT4-mini-Chatclient. This allows prompting Monster-Descriptions and obtaining generated texts. For Image-Generation I use a Dall-E-3-ImageClient, allowing generation of images.

Structured Output

Being one thing that fascinated me the most, I tried adding interpreting the text provided by the ChatClient as a class. For doing that, I needed a sufficient complex datastructure, so I implemented a DnD Monster. Using the MonsterController, you can prompt to create a Monster and returned it according to the Monster.classstructure. This works very well but comes with some minor problems: Sometimes values are not set. Sometimes strings are not uniform between ChatClient-Class (like one time a modifier is '+0', the other time its 'none'), which can cause some errors.

Vector-DB

I added Neo4j to use it as a Vector-DB, since it provides distance-searching using Vector-Indexes. However I currently only persist generated creatures and objects there, I don't use the Vector-Index so far. In the future, this would allow implementing an update-process for existing monsters, adding them as "correct" monsters to the DB. Using Vector-Indexes would allow to more finely orient the generation on monsters you like.

Next Steps

• I want to add data to a simple VectorStore using it for RAG.
• I want to look into structured outputs to return results as structured Objects.

• I want to persist generated data to a Neo4j database.
• I want to add a frontend.
• I want to use the Neo4j as a VectorStore, using the already persisted (and maybe favorited) entities for RAG.
• I want to look into FunctionCalling for calling APIs from inside the LLM, maybe implementing a Combat-Simulator to double-check the Challenge-Rating.
• Maybe I add Artifacts, NPCs, Player-Characters and Places too. We'll see...
• ???
• Profit

License

This project is licensed under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0).