Want to get my receptionist in production. What would help for all of us:
- A sandbox to test what we’re building, not at the individual stage level, but in basically a production level sandbox.
- The 500 character “prompt” (ie, company description) is way too short. It also may be better used as a traditional system prompt rather than an faq generating action (“You are a helpful knowledgeable and kind assistant for company XYZ… Always use {variables}. Never say {variable}. Do not insert stock phrases like “would you like me to transfer you to X”. Trying to build a capable assistant not a switchboard.
- Please provide more information on the knowledge base usage, and how the AI references it.
we can create (some of us have this handy) our own faq; can the knowledge base contain instructions on bot behavior…tone/temperature etc? (More than the options for the default voice and style)? - Will growing knowledge bases create significant latency in the AI response?
- are .md or .xml files usable in this feature? Those seem to be the standard with many agentic systems (as well as the current RC IVR config file format).
- Are key value pair (FAQ for example ) lists better in a csv file or a text/pdf file?
- An off ramp for non English callers. There has to be a way for the agent to recognize non-English, and have its own path.
- What benefits/limitations exist if the knowledge base contained a list of URL’s (that basically link to another knowledge base that relates to this one? Can the current version of AiR perform a real time web search like openAI, Gemini and Perplexity?
- it would be helpful to know the underlying infrastructure of the RC AI build.
I can go on.. but if anyone out there has any answers advice or creative workarounds, would love to hear them!
