Bearth.ai

Done by: Anthony Yip (ayip2) Christian Ang (cang2) Jialing Bi (jialingb) Janessa Guo (janessag)

Inspiration

Bearth.ai generates hypothetical scientifically-based planets. It supports the following primary objectives:

• Enable artists to create scientifically plausible fictional worlds
• Spark interest in STEM research through self-directed (planetary) discoveries
• More efficient planetary discovery by predicting existence/location of new planets

What it does

• Optional user input: Stellar Radius, Stellar Mass, No. of Stars, Orbital Period, Planet Radius
• Extrapolation of features (predict): Star Metallicity, Planet mass
• Identifies likely star system
• Generates visualisation for planet and star
• 3 fun facts about planet based on user input

How we built it

• AI models (tabgan, SGD) for extrapolation of features, and identification of probable star system
• OpenAI API for visualisation of planet and star, and fun facts about planet
• Front-end app using Kivy

Challenges we ran into

• GANs: Establishing correct scalars and transformer for pipeline
• Classifying Star Systems: Too many categories prompted us to use a regressor to bootstrap a classification
• Prompt Engineering: SDM does not understand raw data
• Learning many new technical skills and about the theme of space, and creating a project within a short time span

Accomplishments that we're proud of

In 1 day, we built a functioning minimum viable product. We also acquired and exercised several new skills, such as...

What we learned

• Training a generative adversarial network
• Stochastic Gradient Descent
• Calling on openai API
• Front-end mobile app development
• Creative problem solving

What's next for Bearth.ai

More data sets would allow us to increase reliability in predicting and discovering new habitable planets.

Built With

• generational-adversarial-network
• kivy
• openai
• python
• stochastic-gradient-descent
• tabgan