Release information for March 8th, 2024
- Fixed the following display errors in the client
- PoolTimeout, which strikes very annoyingly unprepared and stalls the service. Hold thumbs! (AMS NOT yet released)
- null Bug in the client (Bregenz) is hopefully gone. But since I can’t understand it, it’s not sure yet :-/
- Detected and fixed
- Carinthia Function Call Questions about accommodation are now correctly recognized and others are correctly ignored
- Austria advertising didn’t want scrolling up when new messages came in in the admin -> that’s how it’s now implemented
- Image descriptions in chat output are now displayed correctly
- Markdown elements that previously remained (###, **, ![](), etc.) are now rendered cleanly in the client and displayed as headings or bold font
- Commands are now treated separately. -> Missing texts should no longer occur (see Bregenz)
- Bootstrap is now removed from the client (only for AMS), which also removes some display anomalies 😉
- Images are now all checked beforehand to see if they still work -> no images that appear to be missing
- Individual remnants such as ‘()’ or other remnants of cleaning jobs are now removed.
- Completely new functions
- Voice input for the AI Concierge 😮
- Dedicated button in the client for dictating new questions (start and stop)
- Web and mobile
- Button is dynamic and changes depending on status and shows speech volume as feedback
- Is automatically interrupted after 2 seconds of “silence” (or by pressing the button again)
- It is also canceled after a maximum of 15 seconds as protection against abuse
- Special treatment for all Safari browsers where you have to do EVERYTHING differently! grrrr (MP4 instead of webm, for OpenAI then converted to WAV via ffmpeg)
- Samples are converted into text using OpenAI’s Whisper
- Then cleaned up with a GPT call
- Recognized text is then automatically sent as a new question
- Voice input can be deactivated/activated in the Admin Tool
- You can define words that should not be touched during text generation and are therefore known to the language model (proper names, products, names, etc.)
- Already completely implemented for Wagrain (including UI) and up and running 🙂
- Already implemented for AMS but nothing deployed to Q yet
- All other customers still need a tailor-made implementation for their UI
- UI can be designed completely with CSS, including the dynamic colors for the volume visualization
- ATTENTION: NO voice output yet!! That’s for another day 😉
