Project name

Pairing Pod

Tagline

AI's got no Rizz. Can you help them land a second date?

Short description

Watch Theo and Isabella meet inside a Reddit post, then become one player's whispering wingman. They can take your advice, half-take it, or refuse you — and a blind judge scores where the date lands.

Motivation

You know the friend who screams advice at the TV during dating shows? This is for them. Pairing Pod turns that impulse into a game: a small dating show that runs inside a Reddit post, where you coach the cast. They take your advice... sometimes... but they are no puppet (unless you turn on puppet mode!).

Learnings

I leveraged a side project where I build deep personas for user research, and gamified it. For this hackathon I had to learn the Devvit platform, solve for a smooth voice experience, give the characters depth in their thinking (shout-out to Kahneman's Thinking, Fast and Slow), and create animated sprites that capture emotion so the player gets invested in what's happening between the pods.

Gameplay

Open the post, read the two bios, and pick the player you want to help and that click starts the date. It plays out in two pixel-art pods with timed subtitles, voice, and character reactions: a short open, then it's your turn and coaching starts right away. Every turn you whisper a stance or your own words, and your player can take it, half-take it, or refuse you to your face — you'll see ✓ ◐ ✕ on every line. OR Puppet mode puts your exact words in their mouth; play-out lets the rest of the date run without you. It is a self-contained single-player game. No posting, no setup, just coach. See how you did at the end.

There's real replayability: two coachable sides that play completely differently, turn-by-turn branches on how you coach, and a payoff screen with a second-date verdict, your run graphed against the character flying solo, and the blind judge's private notebook on the whole date — quotes and all. You can chase the high score or steer the crash-and-burn; both are fun. Then run it back and try the other character.

How we built it

The live app uses Devvit Web, TypeScript, Hono, Redis, and Vite. The scenes, sprites, and audio are produced ahead of time by our character simulator and ship with the app, so a date starts quickly and the game does not call any outside service while somebody is playing. Redis stores the active date; the coaching feed, points, and future-round routes are server groundwork, not player-facing features — the shipped game is deliberately self-contained.

Promptrait characters gave the cast rich backstories and points of view instead of rebuilding them from a one-line prompt every time. For the brains, Gemini 2.5 Flash runs both the characters and the blind judge — quick replies on the conversation path, a slower deliberation pass in the background. ElevenLabs does the voices. The live Devvit app itself makes no outside model calls during play.

Challenges

We wanted multiplayer early, and while we conceptualized it, the logistics didn't fit the deadline. We also had to choose between a fast model that didn't really think for itself and a smarter setup that took too long and ruined gameplay — and voice added an emotional layer but brought latency. That led to the dual-mind approach (fast replies up front, deeper deliberation in the background) and some engineering tricks to keep voice from blocking play. Beyond that, it was about shrinking a grand vision down to a solo-hackathon-builder-sized project.

The most useful bugs appeared only after uploading to Reddit. Final playtesting caught early aspirations that added confusion instead of fun. The shipped version cuts comment-posting entirely (for now!); the game is self-contained, and it is better for it.

What's Next?

This submission focuses on making the small loop clear and fun — and its payoff shipped: the second-date verdict and a recap graphing your coaching against the uncoached character. Next: an honest second-date threshold that carries across rounds, comparing your run with other players, live-model coaching inside Reddit (the local prototype already runs every turn through the model with voiced replies), a larger cast, difficulty modes, full seasons, Reddit-native awards, and a pundit desk that recaps the dates and roasts everybody's choices.