Welcome to PiNeurons! Ever wanted to build your own personal AI to showcase to your friends? Now is your chance!

Design a custom AI model from scratch, and we'll reward you with a $15 OpenAI credits grant or ship you a Raspberry Pi Zero 2 W to run it on! In this project, you'll learn how to build your very own character-level Recurrent Neural Network (RNN) that generates text based on a "personality" you give it.

What is PiNeurons?

PiNeurons is a YSWS (You Ship, We Ship) designed to introduce you to the core principles behind modern AI. While everyone is using tools like ChatGPT, very few understand how they actually work. PiNeurons bridges that gap by guiding you through building a small-scale language model from first principles.

Your model will be trained on your computer and is perfectly scoped to run on lightweight hardware (like your new Raspberry Pi!).

What You'll Learn

By the end of this project, you will understand:

• Data Representation: How text is turned into numbers for AI to process.
• Sequential Learning: How neural networks learn patterns over time.
• Data Quality: How the data you feed an AI shapes its behavior.
• Hyperparameters: How tuning different settings changes your model's performance.
• Text Generation: The magic behind how AI predicts the "next word".

Project Steps

We have two reward tiers based on the time you invest in active development on this project:

• 3+ hours: Build and train your core AI model to get a $15 OpenAI credits grant!
• 6+ hours: Go the extra mile by building an interactive app for your AI to get a Raspberry Pi Zero 2 W!

You can always spend more time if you want to dive deeper!

1. Dataset Engineering

You'll start by building a custom dataset. This decides your AI's "personality". You can choose themes like:

• Storytelling 📖
• Motivational content 💪
• Humor/Meme generation 😂
• Horror or suspense 👻

Requirements: 50–150 KB in size, consistent tone, and clean formatting. Quality data equals a quality model!

2. Model Training

Using our provided codebase, you will train a character-level RNN on your dataset. The model learns by predicting the next character in a sequence, just like large language models do! (Tip: You can find an example Python script to help you train the model inside the /example directory!)

3. Hyperparameter Exploration

Experiment with your model's settings to get the best results:

• Context window (Sequence length)
• Model capacity (Hidden layer size)
• Number of layers
• Output randomness (Sampling temperature)

4. Output Quality Optimization

Your final AI must generate coherent text!

• It should contain recognizable English words.
• It must reflect your chosen theme.
• It needs to maintain readability across 2–3 consecutive sentences.

5. Interactive Running Code

Besides having the model trained really well, it should be interactive in some way. You can build a CLI tool, a React app where you can interact with it graphically, or anything similar to bring your AI to life!

How to Submit & Get Your Rewards!

Once you have a working AI that meets the quality standards, it's time to submit!

1. Create a repository on your own GitHub account and upload your whole project there. This must include your AI Model, Training Code (including dataset!), and Running Code (the app that makes the model interactive).
2. Fork this repository (the PiNeurons repo) and create a new branch.
3. Inside the /submissions folder, create a new folder. The folder name must exactly match your model file's name (without the .pth extension). For example, if your model is MyAwesomeStoryAI.pth, the folder should be /submissions/MyAwesomeStoryAI/.
4. Inside your new folder, upload only these two files:
   • Your trained model file (.pth extension).
   • A manifest.json file configuring your submission.

manifest.json Template

Make sure your JSON file looks exactly like this, filled with your details:

{
    "name": "MyAwesomeStoryAI",
    "description": "A creative AI that generates short sci-fi stories.",
    "thematic": "Storytelling",
    "image": "https://link-to-your-image.com/image.png", 
    "sequence_length": 50,
    "repository": "https://github.com/your-username/your-awesome-repo"
}

(Note: The image field is optional, but if provided it must be a URL to a 1024x1024px image representing your AI!)

5. Create a Pull Request (PR) to this repository.
6. We will evaluate your submission based on functionality and output quality. If accepted, we will send your $15 OpenAI credits grant or ship your Raspberry Pi Zero 2 W! 🚢📦

Happy building, and we can't wait to see what you create!