Dead Ringer

Game Page

Inspiration

We all have love for horror and mystery genres and we thought it'd be interesting to challenge against an AI to solve and think critically against it

What it does

Dead Ringer is an interactive chatbot game where you have a limited amount of messages to uncover clues about which potential criminal you could be talking to. Throughout the messages, you need to critically think about what they say as well as what you ask. Throughout these messages, you will fill out a potential clue board of who the supernatural creature could and other qualities they hold.

The randomly chosen criminal, powered by Gemini AI, sends the first message, along with a voiceover using ElevenLabs.

The user can respond by talking verbally using ElevenLabs’ STT or text input. Gemini AI receives the user message back for the next reply. Messages from the criminal are extracted for any clues that lead to their identity. Clues are saved to a clue board for users to reference when making an accusation on who the user thinks the criminal is. After 10 turns or a manual accusation, users can fill out information about the criminal to receive a score.

How we built it

Frontend: React + CSS AI: Gemini API for Character dialogue and clue extraction, ElevenLabs for Conversational agents and Voices (TTS & STT + Scribe) Backend: Next.js Deployment: Vercel + .TECH domain name

Challenges we ran into

At the beginning we ran into issues where Gemini API had usage problems. The free credits we were provided had an issue configuring and ran into errors where they were at “max” credits. During our building process we also would run into the issue where the AI would often hallucinate . It would give key information away, often confusing the player’s goal with its own. We created a fallback solution in the case where instead of using the players name it would refer to them as the detective, so it would decrease its chances of hallucinating itself.

Accomplishments that we're proud of

We are proud of being able to work with AI agents that we implemented under character descriptions, text to speech, speech to text, as well as extracting clues based on conversation history. In addition we also proud of designing an interactive, fun product that satisfies the itch in your brain when you’re feeling bored. It truly is fun to converse with the AI bot and see how far you can push it. Moreover, we feel incredibly accomplished to have published it on a domain.

What we learned

We learned to implement API keys in order to create a chatbot that We learned how to adjust AI bots to avoid hallucinations and stay consistent with their prompts and selected voices from ElevenLabs. Applying microphone for STT from ElevenLabs was Project management using Trello and Miro and Github helped a lot with keeping track of tasks and writing down ideas out especially during a 24 hour hackathon

What's next for Deadly Ringer

In the future we want to implement better scalability, multiple AI agents where only one is a killer and you must be mindful because only one is the criminal. In addition, we also wish to add a version where it simulates a phone call and you have notes on the side of the audio conversation going on. We also hope to add more social aspects such as a leaderboard! We want to continue challenging others to think critically when it comes to interacting with artificial intelligence, through these simulations/games.