HN Showcase
Back to browse
Ball 2 – Use AI to build fun ball games

Ball 2 – Use AI to build fun ball games

by klangdon·Feb 19, 2026·2 points·3 comments
Visit ProjectView on HN

AI Analysis

●●●BangerZero to OneCrowd PleaserWizardry

Sensor-packed foam ball with AI game creation—solves the "what should we play" problem.

Strengths
  • •Hardware + software combo genuinely novel: embedded IMU in durable foam with 500+ hour battery life
  • •Voice-first game creation via OpenAI Realtime API removes friction for kids and non-technical users
  • •Addresses real pain point: outdoor/active play plus creative expression without typing
Weaknesses
  • •Hardware distribution and scaling are non-trivial; unclear if unit economics support $69 price
  • •Relies on OpenAI Realtime API; game quality depends entirely on LLM consistency and creativity
Category
Hardware
Target Audience

Families, kids, educators, and anyone wanting physical play with AI-driven game creation

Similar To

Sphero (connected ball toys) · Roblox Studio (user-generated game creation) · Nintendo Switch Sports (motion-tracked physical play)

Post Description

I wanted to find a way to get kids:

- moving

- creating

- playing with others

In the end I, and a team of 4 others, built:

- a foam ball with a BLE IMU in it (moving/together)

- an app that allows you to create almost any game you can imagine with AI (creating)

Video of teenagers making a game: https://www.youtube.com/watch?v=-7TZXRBybOE

Video that was intended to be for Kickstarter: https://www.youtube.com/watch?v=Edy9zew1XN4 (we abandoned kickstarter, long story)

For hardware we came up with a pretty awesome custom PCB, based around a Dialog DA14531, that doesnt need any charging for over 500 hours of play. The electronics are encased inside 4.5" diameter of foam. We drove a car over it and it didn't break.

For the app, we are using OpenAI Realtime API (gpt-realtime-2025-08-28) to gather the game requirements via conversation. We are using voice/realtime to allow kids to do it without needing to type or read. This piece has a huge dynamic prompt that flows with the conversation. It has about 20 different tools that the agent can use to access sample requirements, ball data, user profiles, api documentation, etc. We also did a lot of work to require younger kids to have adults present and make sure we never saved any pii, etc.

Then we use Gemini 3 Pro to process the conversation and generate a markdown specification/arhictecture of how the game should be designed. We found that Anthropic Opus 4.6 did perform better here, but Gemini 3 Pro is much cheaper and faster. This has a static/cacheable prompt that is primarily a ball api documentation and details on previously seen issues.

Then we use Gemini 3 Pro to also code the game. A very similar prompt to the specification/requirements just different purpose. We are currently exploring condensing the architecture and coding steps into a single step, with new "thinking" models. Also seeing that Google is launching 3.1 model, literally while I type this...

The end result is an app that allows you to speak game requirements, have ai code the game, then you get to play it with a real physical ball.

So ya, tons of fun building all that. Curious what HN thinks. Now the hard part of maybe getting someone to buy it...

Similar Projects