Build AI Support Reply Agent

I’m putting together an AI-driven workflow that will draft the first-pass technical support replies for every new case our customers open in Salesforce. The agent must pull the relevant knowledge-base articles stored in Confluence, extract the key points from those pages, and inject an accurate, context-aware response back into the Salesforce console so my team can review, tweak if needed, and send.

Here’s what I need from you:

• A robust link between Confluence’s API and an NLP pipeline that keeps an up-to-date index of all knowledge-base articles.
• A key-point extraction model (LLM / embeddings or your preferred architecture) that can surface the most relevant snippets for any incoming case.
• Tight Salesforce integration—Apex, Flow, or REST—so the suggested reply automatically appears inside the case record, ready for agent approval.
• Clear evaluation metrics and a quick feedback loop so we can iterate until response accuracy, tone, and coverage meet acceptable thresholds.

If you’re comfortable combining Confluence APIs, modern language models, and Salesforce development to create reliable technical support automation, let’s talk details and timelines.