🎓 Multi-Part A - Z Buildout For Our Support Voice Button

Jan 9 (edited) • 🏰 Homebase

Part 1 - Prompt Engineering & Conversation Lab Testing (1:04:56 watch time):

Part one is a full prompt engineering walkthrough of building our multi-protocol voice AI prompt for a programmable button widget. As I prompt, I show how I use the voice lab to test the conversation to refine key areas of the prompt. I try to show different prompt engineering techniques like Identity building, Personality traits, Task framework, Protocol building & referencing, conversational flow building, etc.

Part 2 - Knowledge Base Upload & Document Creation (0:09:01 watch time):

Part two is using the 3rd party supabase RAG system on buildship, I didn't go through that process since there is already a video on it but I set up the environments on camera. Along with that, I show you how to use your voice to create a parsable document for your AI to upload so you dont have to write documentation to be uploaded and I show uploading and testing the RAG system.

Part 3 - AI Programmable Button Creation & Testing (0:28:51 watch time):

Part three is the creation and testing of the button widget that I am going to use for this deployment by putting it in a GoHighLevel funnel page and testing live while refining the RAG system and prompt. This part includes a lot of testing and prompt refinement based on the widget's answers and why I make the changes I do.

Part 4 - Maintenance Setup, Deployment Refining & Closing Remarks (0:24:33 watch time):

Part four is setting up a post-call webhook to collect the data in a 3rd party database (I use a buildship database for this one) to be able to easily maintain this deployment overtime. I also go through a few more points of refinement of the widget, prompt, and tool calling to ensure the answers are accurate and the conversation flows well with autonomous decision making and abilities. And I end it off with some closing remarks on why I think understanding some of this stuff deeper will keep you competitive in future deployments as AI becomes more commoditized.

------

🤖 → Copy bot by ID as a reference: 1736268047820x399108066375019970

ℹ️ → RAG using Buildship & Supabase: https://youtu.be/y5g6zL3_plE

🧠 → Mind map attached for prompt conversation flow

🔊 → Call recording attached for you to hear the question answer ability and conversation flow

14 comments