Inspiration

During the email usage study work in the office, scientists found that 70% of e-mails recipients responded within six seconds after receiving, and 85% within 2 minutes. On average, 28% of work time is spent on email and we check email on average 11 times per hour. Beside social media, email might be the biggest distraction from your real work.

Less email, more productivity

With OpenReplAI you do not have to stop everything what you do when you receive an email, like today.

OpenReplAI is an AI-system that can process long emails and reads back to you the essentials as in an executive summary. No matter if you receive a positive, negative or mixed message, based on the sentiment and key phrases a personalized response will be context generated. Finally, you can easily confirm sending or modify the reply before it is sent.

OpenReplAI solves the time-consuming email problem within a fully interactive system that integrates seamlessly in what you do.

How we built it

Adding speech interaction and text analytics in any application is pretty straight forward and OpenReplAI is using Azure Cognitive Services, such as Speech 2 Text, Text 2 Speech and Text Analytics. link

In order to generate summaries and replies there is the amazing GTP3 model from OpenAI. link

As alternative you can build your own custom model based on your sent messages to summarize and generate emails.

Added all together into an application is done using Visual Studio 2019 link

Challenges we ran into

Sometimes getting responses from the AI model is pretty random, but it get better after some shots, therefore you need to confirm before sending back messages.

Accomplishments that we're proud of

Managed to integrate all the cognitive services pretty fast.

What we learned

We could really make steps in getting back on our important work and automate some of the time consuming jobs.

What's next for OpenReplAI

Distribute into the Office Marketplace.

Built With

  • azure
  • cognitive-services
  • openai
  • text-analytics
Share this project:

Updates