Inspiration
Our team goal was simple; empower others to be the best they can be. One of our team members worked at a performance management startup recently and had a lot of thoughts on how AI can be leveraged to generate objective and timely feedback. Good feedback is critical for professional growth and is, unfortunately, difficult to source. Studies show that women and people of color receive less constructive feedback because managers (who are predominantly white and male) are less likely to have difficult conversations with people they don't feel comfortable with. Moreover, today's managers are overworked and often unequipped to give productive feedback to begin with. We believe everyone should have access to constructive feedback to improve EQ and productivity in order to become the best versions of themselves.
What it does
Eloise is an AI executive coach that gives detailed feedback on soft-skill professional tasks like one on one meetings or presentations. By uploading a video or audio file of a meeting or presentation, the product gives a detailed feedback report of what was done well, what could be improved, and highlights any behaviors or trends that may have been overlooked by the user.
How we built it
We leveraged the hume API to analyze the video and audio files using the empathetic voice interface and expression measurement API to collect information on interpersonal dynamics, presentation effectiveness and tone. We then feed the analysis to OpenAI, prompting it to analyze the effectiveness of the presentation or conversation by summarizing trends or behaviors that may have been missed as well as recommendations for next time. The recommendations and analytics of the recording are then displayed on a web page. Videos and audio files can be uploaded natively in the app or via WhatsApp using a Twilio integration.
Challenges we ran into
- The Twilio integration had quite a few bugs and was not working for a while.
- We came across many errors with the Hume API only to realize the API required much longer videos for the text transcription to work. We had to move from ~30 sec videos to ~2 min videos which then takes a longer time to process.
- We had some issues getting Replit to work since none of our teammates had ever used it.
- Diarization with Hume has proven to be tricky
Accomplishments that we're proud of
- Getting a functional product!
- This was the first time a lot of our teammates used the Hume API and the first time one of them used the OpenAI API. Everyone learned a lot!
- The Twilio integration actually works!
What we learned
- Everyone used Replit for the first time
- 3/4 learned about Hume and how the API works
- 1 person learned how to work with the OpenAI integration
- 2 people did a deep dive around
- 3 people learned a lot about the bias in feedback and the need for a product like this!
What's next for Eloise: Personal Executive Coach for Emotional Intelligence
We'll see once the awards are announced ;).
Log in or sign up for Devpost to join the conversation.