Voiceflow isn't recognizing entities
I'm simultaneously experiencing 2 issues:
- Voiceflow isn't recognizing entities on the first try
- Voiceflow isn't reprompting for missing entities
Screenshot images of the chat and logs are in the attached images.
There are 4 entities:
- knee_side: left knee, right knee, both
- symptom_description: hurts, aches, numbness, tingling...
- symptom_activity: run, walk, squat, bend, straighten...
- symptom_location: back_knee, front_knee, behind_knee...
The prompt: "the back of my left knee hurts when I bend it"
The prompt contains all 4 required entities:
"symptom_description": "ache",
"knee_side": "left",
"activity": "bend",
"symptom_location": "back_knee",
But Voiceflow only recognizes 2 of them on the first try: activity and symptom_location.
Instead of continuing the flow, Voiceflow gets hung up. In the logs, it shows:
Fill "knee_side_entity"
But the chat window doesn't show a reprompt. Instead, it displays nothing.
So I type in "left knee" then the flow continues. It then asks me for a symptom description. I say "it hurts". Then the flow resumes.
All 4 of the required entities were contained in the original prompt. So why is Voiceflow only recognizing 2 of them on the first prompt? And when it needs a knee_side entity, why is it being silent instead of sending a message to the chat window?
5 Replies
Can you share a screenshot of your flow?
With the very limited flow you screenshotted, it looks like you are doing back to back captures. This is why you are having to prompt 2 times. You should really just use an Agent step to handle the whole "entities" stuff. It will be able to capure everything and then continue the flow from there.
My 2 captures aren't in consecutive blocks. You'll see the first capture block in the screenshot as "Save entire reply to { last_utterance }". After that block is a Set block, then a Conditional block, then my 2nd capture block. So my 2 Capture blocks are separated by 2 blocks.
Why are you saying the screenshot is limited? This shows the entire flow so far. The chat ends at the Conditional block on the top right. Would it be more helpful if I showed more of my screen in the screenshot?
Thanks for the advice regarding replacing the capture blocks with an Agent block. I didn't realize they were synonymous. I'll try it out.
Yes, but the set and condition blocks don't show anything. So you are capturing and essentially capturing again. As for being limited, I'm not aware of the path your flow is taking based on your question. Therefore, there might be another path with a different issue.
Again, an agent step would be way better at doing this, IMO.