Speak Smart

Inspiration

Inspired by Hume.ai's unique LLM approach, we decided to build a tool to help others understand their expressions they give off during their conversations and speeches, while also as a developer to better understand Hume's capabilities.

What it does

Given a text input and a list of emotions, it provides a simple UI/UX with Gen AI-generated tags of the text indicate how expressive of each emotion you should to make your speeches impressible. Afterwards it summarizes your results, comparing with the provided guidelines, and ultimately provides a powerful tool for people to routinely practice their public speaking skills.

How we built it

We used FastAPI backend + Next.js frontend. Our sentence similarity ML endpoint was also hosted on an AWS EC2 instance, as we expect higher stability from such infra. :)

Challenges we ran into

We ran into websockets streaming randomly cutting off.

Accomplishments that we're proud of

We are proud of building end-to-end our application, despite having spent much time exploring new APIs at workshops and just learning more about the startups tabling and others' backgrounds and projects.

What we learned

We learned how to prototype end-to-end GenAI products using multimodality like speech-to-text for the first time, some CSS styling tricks.

What's next for Speak Smart

• Enhanced UI/UX through more user surveys
• More robust end-to-end integration
• Smarter metrics created to optimize efficiency and reduce variance in our outputs
• Explore new features such as video integration of famous speeches and their speakers to more effectively learn how to become a better speaker

Built With

• amazon-web-services
• next.js
• openai
• python