Inspiration
Inspired by the classic allure of whodunit tales and the capabilities of PartyRock, I was interested to see if I could create a Murder Mystery app where each mystery unfolds uniquely, based on a theme entered by the user. This project is a testament to my love for mysteries and AI possibilities and has resulted in unique interactive mysteries to solve.
What it does
The user enters a theme which generates a short story based on that theme with a crime to solve, a victim and several suspects, each with their own AI-crafted personality. The user can interact with each suspect via their own chatbot, providing clues, red herrings, and alibis, ensuring every story is not just a reading experience but an interactive investigation.
How I built it
I used the PartyRock app, an Amazon Bedrock Playground to learn about Prompt Engineering and Foundational Models (FMs) and built a functional app using generative AI. In more detail, I used Claude to generate the story based on a user-entered theme. The prompt given, ensures each story has a crime scene, a victim and three suspects. Claude was also used to generate a description of each of the suspects based on the generated story. I used Stable Diffusion XL to generate images of the crime scene and the three suspects. Finally, I created four chatbot widgets, which are designed to take on the personalities of the the 3 suspects and the Detective’s Assistant whose tasked with helping the user solve the crime.
Challenges I ran into
At one point, after I changed the prompt, Claude stopped returning output as it deemed it could be sensitive, so I had to change the prompt to make it clear this was a fictional story. Also, coming from a development background, I found it difficult to not think in terms of variables and parameters and struggled at first to work out how I could set up the unique chatbots. I was able to overcome this by creating widgets for each suspect and using these in the prompt for the chatbot(s).
Accomplishments that I am proud of
I was surprised the idea worked so well and it was great to find that my fine-tuning of the prompts, made the app more robust. Having tried various themes as diverse as Ancient Rome, Victorian England and 25th Century New York, it was great to read the different stories and how well the characters created blended with the theme.
What I learned
I have gained a good understanding of prompt engineering and how to fine-tune prompts to get a more specific output which was crucial to give each chatbot their own character. Since, creating the app, I have used my prompt engineering skills on more business-related cases.
What's next for Murder Mysteries
I would like to investigate the possibilities of longer stories and more intricate tales and, of course, more devious characters!
Built With
- amazon-web-services
- partyrock
Log in or sign up for Devpost to join the conversation.