-
Notifications
You must be signed in to change notification settings - Fork 3
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
speeding up responses #75
Comments
I haven't done anything with this project for a while, have been focusing on Chime vs Connect. In this repo the Lex bot does have to choose between some intents and the output back to Connect is an intent like transfer for example. In my new project I removed all that so fallback is essentially its only option. That does seem to speed things up. I literally rewrote the GPT code for speed and using a function model to return and trigger call control activities (as well as switching languages on the fly). For Both AWS Chime SDK and AWS Connect the only real way to get anything close to a conversation requires streaming voice, and AWS Lex is the only viable integration point for both AWS services. Take a look at: https://github.com/docwho2/java-squareup-chatgpt-ivr which is light years ahead of this repo in terms of functionality on the ChatGPT side and does multi-channel, Facebook conversation handover, truly multi-region with failover, etc.. The connect stuff was a simple POC, the ChatGPT IVR is for real and deployed and running in a production setting and the feedback has been amazing from both customers and AWS folks. Actually meeting a reporter tomorrow for a newspaper article. Try calling +1(612)254-0226 which is just a sandbox deployment / test number and let me know what you think about how the conversation flow is compared to what you're seeing with this project. The models are pretty fast now compared to what they were before, so once the initial lex bot is fired up most people relate that it's fairly close to conversational. I did allot of Lex tuning in terms of how it responds, for example if GPT can return response within 3 seconds, you simply hear the response, after 3 it will say that its thinking/processing, etc. And I use GPT's responses in all cases vs any kind of static prompting when ending a call for example. In general focusing on making things feel more conversational and natural. I was just at a trade show speaking to a woman from Sweden, she conversed with the bot over the phone in Swedish and asked all kinds of questions about our store and she was blown away. Would love to hear your feedback using the test number mentioned above, Steve |
Is there any way to either 1. keep continuous dialogue going w/out a lex bot?, or 2. immediately trigger a fallback intent for lex? This is a fantastic project, but responses are extremely slow and break the 'feel' of a real conversation. Thanks for sharing it openly!
The text was updated successfully, but these errors were encountered: