Transform

• 

  Logo

What

We built a full-stack application built for education, while leveraging generative machine learning models (GPT-3, DALLE-2) and foundational machine learning models (Resnet-50). The website is developed around 3 main sections- Art, Finance, Programming.

Art

We use DALLE-2 to create an art challenge. You prompt an image for DALLE-2 to create, and your goal is to re-draw it.

We score how well you matched the original image by utilizing the embeddings of both images, generated by a pre-trained Resnet-50. We then used a cosine-similarity calculation to calculate the angle between the 2 high dimensional vectors- by comparing the angles of the 2 vectors, we can determine how "similar" the 2 respective images are to each other, since the embeddings of the network contains semantic information.

Programming

We use GPT-3, and special prompt injections, to allow code translation, and code explanation. Simply paste your code and either ask for translation into any other programming language, or ask for code explanation in natural language

Finance

Based on several inputs about your financial state (income, age, financial goals), we prompt GPT-3 to give you responsible financial tips based on that state! We made sure the prompt was injected in a way that GPT-3 would not give irresponsible financial advice

Built With

• django
• pytorch
• react