ChatGPT Support Chat Automation -- 4

I’m building a self-serve, AI-driven help desk and need a specialist who can turn ChatGPT into a friendly, always-on support agent. The focus is narrow and clear: generate accurate, conversational chat responses that feel genuinely helpful rather than scripted.

Here’s what I have so far:
• A knowledge base containing product documentation and common troubleshooting steps (Markdown and PDF).
• Access tokens for OpenAI and our live-chat platform (Zendesk Chat, but I can switch if you recommend a better fit).
• A rough list of recurring questions pulled from the last six months of transcripts.

What I’m missing—and what I’d like you to deliver—is a refined prompt-engineering workflow that teaches ChatGPT how to:
1. Detect intent from incoming chat messages.
2. Pull the correct answer (or the most relevant snippet) from the knowledge base.
3. Respond in a consistently friendly tone, adding light empathy and clear action steps.
4. Ask clarifying questions when needed rather than guessing.

Acceptance criteria
• A reusable prompt template (or short chain of prompts) that produces high-quality chat replies 90 %+ of the time in a test set of 100 unseen queries.
• Integration code or no-code setup that plugs the workflow into Zendesk Chat and triggers within 1-2 seconds.
• A quick reference guide so my non-technical team can tweak the system without breaking it.

Tools that will probably come into play: OpenAI API, Python (or a no-code alternative like Zapier/Make), vector search for the knowledge base (Weights & Biases, Pinecone, or whatever you prefer), and basic analytics to track response quality over time.

If you’ve already built something similar—or have a clever prompt structure that nails tone and accuracy—I’d love to see a short demo or example output when you respond.