Inspiration

Our team grew up playing many different RPG's and enjoyed text-based RPG's as well such as Kingdom of Loathing. Hence, creating an entertaining application using PartyRock was something that called out to us. When coming up with genre of game we would like to develop, we swiftly came to the conclusion that having multiple different chatbots, each with their own roles, personality and goals was an intriguing concept and we built up this app from there. We thought that the theme of a council/courtroom filled with court intrigue and individuals/parties with their own intentions was a great fit to this idea. Hence we came up with Mechs & Magic, A game where you have to manage the expectations of both wizards and inventors which have very different goals and mindsets and your decisions affect BOTH factions. For the story, we drew parallels from the real world where the two parties represent the dilemmas we often face when juggling innovation/Technology/Advancements versus Environmental concerns.

What it does

The game is a text-based RPG game where the worlds setting is communicated through an introduction at the top of the application as well as some images. A shortened version of the introduction is that you are a counsellor in a city that is facing a pressing energy crisis. There are two major parties, the Wizards and The inventors who have different views on solving the energy crisis. You must make your decisions and type it into the user input field provided. These decisions will affect your relationship with the different parties and you must manage these relationships carefully. Your current statuses and relationship levels are tracked in a log provided in the application.

How we built it

We spent quite a bit of time generating ideas and refining the story for the world we wished to portray. After which we engineered the prompts for each character individually one by one to ensure that the chatbot gave the best responses. In total the application includes 3 AI Image Generators, 5 AI Text Generators, 1 Static Text and 1 User Input Field. During the building process we experimented with different text models as well as different Image generation options/presets and after much testing we found a set of options and models that fit our application the best.

Challenges we ran into

The main issue we faced was with the chatbots, mainly we had to make the chatbots not speak like chatbots and instead speak in a first person perspective. This included removing preamble such as "I understand my role, this is what I would say in this role...". Another Issue was getting the Chatbot to act as a log for the world state. At first the AI outright refused to act as a log and insisted it was a Chatbot. Following some testing and changing of the prompts we managed to make it work. Lastly, one major issue is getting the AI to respond dynamically and realistically to a players impute. The AI would often keep repeating the same responses and reiterating the same points despite very different player inputs. The AI would even disagree with the player even thou the players decision was exactly what it wanted. This was largely because a large part of our initial prompt for the AI was the setting for the world, hence most of its replies would be based on what was mentioned in the worlds setting. We attempted to fix this by adding in requirements to the chats response like "Please base your response heavily upon player input only" and we feel it did mitigate the issue to an acceptable amount.

Accomplishments that we're proud of

We are very proud of the story that we came up with especially that we managed it incorporate a theme that is very relevant and important to us, namely technology and innovation and the implications it can have on our climate and natural resources. We also are very proud of getting many different chatbots to work together and that all of these chatbots are relatively in sync and reply dynamically and accurately to the players input. Lastly, even thou its a small achievement, we feel that the images being generated are very representative of what we wished to convey and we actually spent quite a bit of time experimenting with the prompts and styles for the image generation.

What we learned

We learnt that AI chatbots are actually much more advanced and have a lot more functionality than we previously thought of. Before this project, We usually used the chatbots mainly for idea generation for school work or for coding. The chatbots actually taking up various roles even a logging board was a real eye-opener for us and truly excites us to think what AI can become with more time and development. We also learnt a lot more about prompt engineering through this project as we had to experiment and see what works best for us in our given context, we felt that stating the instructions and roles to the AI clearly before moving on to giving context to the AI was a key learning point and also to keep repeating key instructions throughout the prompt to ensure the AI does not ignore it.

What's next for Mechs & Magic

We feel that we can continue to work on the way the AI responds to Player responses. We also feel that we can expand the story further to include many more scenarios and events and also implement some form of progression although that maybe difficult as the AI tends to forget events relatively quickly. Expanding the story and the world and fleshing out the prompts will generally be our main goals moving forward and we are thrilled to see where this project takes us next!

Built With

  • chatgpt
  • claude
  • partyrock
  • stablediffusionxl
Share this project:

Updates