Jest Guessing

TL;DR I created a character guessing game based on a popular YouTube channel.

Both characters have a missing limb or prosthetic

Both characters are bald

Both characters are royalty

Both characters have the same name

Both characters are cyborgs

Both characters are dads

Both characters wear an iconic mask

Both characters have a missing limb or prosthetic Both characters are bald Both characters are royalty Both characters have the same name Both characters are cyborgs Both characters are dads Both characters wear an iconic mask

Background

Recently when I was visiting my brother Daniel in Texas he introduced me to a phenomenal YouTube channel called The Best in Jest. The premise is simple: 2 people get assigned a character and they have to figure out who it is. I love the concept and thought it would be great to be able to play this type of game with friends during a dinner party or at the pub.

What inspired this all: The Best in Jest Youtube channel

HMW create a character guessing game that I can play with friends?

 

Process

Determining the rules

  • Every game has a commonality that all characters need to align to

  • Each player is randomly assigned a character without them seeing

  • Each round, players can do the following:

    • Ask 1 yes or no question

    • Guess their character

  • Once one player guesses correctly, each player has one more chance to guess

Collecting the Data

To give my game agent a reliable foundation, I needed two things: a curated list of fun commonalities and a deep pool of recognizable characters to pull from. I compiled these and converted them into clean JSON files, which I hosted on GitHub so I could easily plug them into different prototypes as the project evolved.

List of Commonalities

Top Characters

 

Proof of Concept

Game Agent

To start, I wanted a simple way to set up the game to provide the commonality and the characters to each of the players. It was a great starting point to understand what my later app would need in terms of functionality. It didn’t work particularly well with more than 2 players but it was a start!

Setting up the agent

I created my agent for free using Gemini’s free feature called Gems (now called Skills). The premise is giving it a set of core instructions and providing the knowledge documents that I created. I stress tested the agent with my husband, playing many rounds and tweaking the instructions until it performed properly.

Jest Guessing Agent set up using Gemini

 

Prototyping

1st Iteration: Stab in the Dark

Full transparency, I actually made this prototype before I created the agent which is why this is fairly unusable. I hadn’t provided actual data sources and my instructions were unclear. Aside from that, I was relying too heavily on AI which meant that I was burning through tokens with not much to show for it. This was made with AI Studio and tried to focus on the idea of people playing around a table.

1st iteration of the Jest Guessing game

 

2nd Iteration: Refining the Flow

Taking my learnings from building the agent, I approached this prototype quite differently. This time I focused on reducing the complexity and solving for latency issues from the beginning. Removing the single player and custom commonality features let me focus on the core gameplay.

2nd Iteration of the Jest Guessing game

 

3rd Iteration: Applying Styling

After solidifying the flow with the second prototype, I created a style guide that pulls in some of the medieval futuristic elements that I love about the Best in Jest YouTube channel.

Inspiration

Medieval Futurism pinterest board

Style evolution

Even with a clear style guide and detailed prompts, getting AI to hit a specific visual bar takes patience. It took quite a few iterations to land on a branding direction that met my standards. Here’s a look at how the visual design evolved along the way.

 

Deploying

A major personal goal for this project was taking it out of the sandbox and deploying it like a real product. I connected my codebase to GitHub, deployed through Vercel, and even bought a custom domain name. Here is a quick look at the workflow I used to take it live:

Launch flow across different tools

 

The Final Result

 

Next steps

Now that the core game is live and stable, I have a backlog of features I’d love to re-introduce:

  • Single-player mode (where you play against an AI host)

  • Custom commonalities so players can invent their own themes

  • Multi-device party mode so everyone can play from their own phone

Key Takeaways

If you're thinking about building your own AI-powered side project, here are my biggest lessons learned:

  • Good data is half the UX: My first prototype flopped because I expected the AI to magically know what I wanted. Taking the time to structure clean JSON data and write clear system rules made all the difference.

  • Protect the core user journey (Less is more): Jumping straight into edge-case features like single-player and custom themes split my focus and wasted time and tokens. Nail the core loop first; add the bells and whistles later.

  • Backtracking isn’t failure: Stepping away from the UI prototype to test the logic in a plain-text Gemini Gem felt like taking a step backward at the time, but it actually saved the project. Don't be afraid to stripaway the UI to validate the mechanics.

  • Don't be intimidated by deployment: Pushing code to production sounded daunting, but tools like GitHub and Vercel handled the heavy lifting and got my app live in seconds.

Next
Next

Takeaways from SXSW London 2026