Inspiration
What it does
How we built it
Inspiration
Every Stranger Things fan has a scene they'd change. What if Hopper really died? What if Eleven killed Vecna first? Fans argue about these what-ifs in thousands of Reddit threads and fan fiction stories, but no product lets you actually play one, inside the world, as yourself, with your friends.
We looked at what exists. Roblox has scale but no real story, and it's adding AI on top of old infrastructure. Episode and Character AI proved people love interactive stories, but they don't remember you, and open-ended character chat is expensive and risky. Newer AI game makers produce generic games with no multiplayer and no world you can live in. We wanted to build something AI-native from day one: a story world that gets better the more people play it.
What it does
The Shared Gaming Lab is a web-first marketplace for playable stories. Our first world is Hawkins.
- Create your own character. You're not Eleven; you're you. Canon characters become NPCs.
- Play the episodes. It's a choose-your-own-adventure story with AI-generated branching choices. Episode 1 is free.
- Unlock the open world. Finish the season and Hawkins opens up as a Sims-style multiplayer sandbox. Landmarks mark key story moments, and clicking one lets you replay or rewrite that scene.
- Play together. Send a friend a link and they join your world. There's no download, no console and no app store.
How it learns
Every choice feeds two loops:
- Personal loop: the game remembers what you chose and shapes your next episode and your world.
- Crowd loop: we track which what-ifs players gravitate toward, and those are the branches we generate next.
For a choice point $c$ with options $o_1, \dots, o_k$, the crowd signal is simply the share of players who picked each option:
$$ p(o_i \mid c) = \frac{n_{c,i}}{\sum_{j=1}^{k} n_{c,j}} $$
The branches with the highest demand get expanded first. As the number of players $N$ grows, these estimates sharpen (the error shrinks roughly as $1/\sqrt{N}$), so every new player makes the story better for the next one.
How we built it
- [Frontend / framework] for a web-first experience that works on any device, from phone to laptop
- [LLM / API] to pre-generate branching episodes and NPC dialogue options from a world bible for each setting
- [Database] to log every player choice and aggregate it into the crowd and personal signals
- Hosted at thesharedgaminglab.com
We chose pre-generated branches and select-option interactions over open-ended chat. That keeps characters sandboxed and safe, keeps compute costs low (we generate a branch once and serve it to every player), and gives us clean, structured data to learn from.
Challenges we ran into
- Scope: we wanted episodes, an open world, multiplayer and a marketplace in one weekend. We cut fully open roleplay, which made character responses too hard to keep reliable, and focused on one world done well.
- Safety and liability: free-text chat with AI characters is where products like Character AI ran into trouble. Select-option interactions let us keep the magic while keeping agents contained.
- IP: we name worlds by setting (Hawkins), not by franchise, and our goal is original what-if storylines inspired by the source, not copies of it.
- Compute: live generation for every player doesn't scale. Pre-generating branches and spending generation only where the crowd shows demand does.
- Hackathon reality: multiple AI sessions running at once, losing progress when laptops closed, and a laptop sitting at 11% battery while plugged in.
Accomplishments that we're proud of
- An end-to-end flow from the marketplace to character creation, an episode, the unlock and the open world
- A learning loop that is core to the product, not bolted on
- A clear platform vision built around one polished world
What we learned
- Curated quality beats volume. Twenty worlds done extremely well beat 20,000 generic ones.
- Removing friction compounds. Web-first means more players, and more players mean faster learning.
- Constraints help. Pre-generated, select-option design made the product safer, cheaper and easier to learn from.
- Looking closely at competitors showed us the gap: no one combines personalization, multiplayer and a world you can live in.
What's next for The Shared Gaming Lab
We see the future of The Shared Gaming Lab as the main marketplace for playable stories, with creators at the center.
- Creator studio for indie authors: upload a book or outline, and we generate the episodes, branches and a world readers can live in. Reader choices then show authors which what-ifs to write next.
- IP collaborations: studios already do this (Roblox × Stranger Things, Episode's licensed stories). Hawkins shows what we'd build for them.
- Community worlds: fans vote on what comes next, starting with Lagos Life, a Sims-style city anyone can live in.
- Smarter agents: world-specific agents trained on fan discussions, long-term memory for each player, and low-cost open-source models. ## Challenges we ran into
Log in or sign up for Devpost to join the conversation.