Lively

Example of prompting Neil Armstrong
Example of prompting JFK with the same question, noting the different generated response
Mobile-friendly user interface, which would be the productionized version
Architecture diagram

Inspiration

We aim to dovetail the power of immersive learning and that of generative AI to build a platform for middle school students to learn about history in an engaging and exciting way. Our project is a conversational AI interface that allows students to hold "real-life" conversations with historical figures. We want to leverage the potential of generative AI (using technologies including Open AI's Whisper, Open AI's GPT-3, GANs, etc.) to "make learning come alive."

What it does

Choose a historical figure who you'd like to speak to 👨‍⚖️
Ask them a question❓
Receive a response with realistic video, voice, and facial movement — as if you're conversing IRL 🗣️
Continue the conversation 🔄

How we built it

See the Architecture Diagram image below.

Used a mixture of foundational models (ie: Whisper, GPT-3, Wav2Lip) as well as custom prompt engineering and full-stack development. Whisper was used for speech (from the user) to text (input for GPT-3). Wav2Lip was used to combine GPT-3 output with videos in order to provide a richer UX experience for students.

Accomplishments that we're proud of

Building two versions of our app (mobile + desktop version). Our lightweight mobile version was built on React/Flask while our desktop version utilized Streamlit.