# Building an Support AI Agent

## **How to Build and Deploy an Internal Support AI Agent in raia**

### 🧭 Overview

An **Internal Support AI Agent** helps your support team get fast answers from company documentation, automate routine questions, and integrate with support systems to log or retrieve tickets.

It is:

* **Trained on your internal documentation**
* **Deployed to Copilot** for daily use and continuous learning
* **Connected to your support systems via Functions or Webhooks**
* **Capable of scoring, summarizing, and notifying based on interactions**
* **Eventually deployed to Live Chat** to serve external users with internal oversight

***

### 🏗️ Phase 1 – Prepare & Train the Agent

#### ✅ Use raia **Connect** to Prepare Content

Upload your support content (PDFs, Docs, Notion exports, etc.) into **raia Connect**, which:

* Cleans formatting, removes artifacts
* Breaks large documents into AI-digestible chunks
* Outputs structured Markdown or JSON
* Pushes content into your agent’s vector store

This ensures your agent has clean, relevant, and retrievable knowledge from the start.

***

#### ✅ Define Agent Instructions

Use **Launch Pad** to define the agent’s behavior, tone, and responsibility.

**Example Instructions:**

> You are an Internal Support AI Agent for ACME Corp.\
> You help the support team answer questions about products, policies, systems, and tools.\
> You respond clearly and concisely using your trained knowledge.\
> If you're unsure, admit it and suggest escalation.\
> You can access ticketing systems via integrated functions.

***

#### ✅ Upload & Organize Training Material

Documents to upload:

* Internal support documentation
* Troubleshooting guides
* SOPs
* Tool-specific how-to docs
* Ticketing process docs
* System architecture or API documentation

Organize content into **Training Packs** for easier future updates.

***

### 🧪 Phase 2 – Deploy Internally via Copilot

#### 🔄 Copilot for Testing and Training

Let support agents use the AI daily to:

* Ask real questions
* Rate responses 👍/👎
* Edit answers and mark “Include in Training”
* Add tags or summaries
* Simulate complex issues

#### 👥 Takeover Mode

In any live conversation, a support user can **instantly take control** of the thread using **Copilot’s human-in-the-loop** mode.

This allows:

* Manual handling of sensitive or complex topics
* Escalation to SMEs
* Seamless handoff back to AI

***

#### 🔌 Connect to Support Systems

Use the following **skills and integrations**:

**🔄 Function Execution**

Set up secure API access to:

* Look up ticket status
* Create new support tickets
* Retrieve knowledge from internal tools

**🌐 Webhook Skill**

Push data from conversations into:

* Jira, Zendesk, Freshdesk, or other systems
* Shared inboxes or support CRMs
* Internal databases

Triggered at:

* `OnMessageEnd` (per reply)
* `OnThreadEnd` (after conversation closes)

**📬 Notification Skill**

Automatically **email or SMS** a human support agent when:

* A conversation begins (e.g., urgent request)
* A conversation ends (e.g., needs follow-up)
* A specific tag or keyword is detected

***

### 📈 Phase 3 – Add Scoring, Summarization & Monitoring

#### 📊 Scoring Skill

Use this skill to:

* **Auto-score** conversations for quality and confidence
* Tag interactions that require human follow-up
* Filter top issues or low-performing responses for improvement

#### 🧾 Auto Summarization

Each interaction can be summarized by the AI and:

* Added to the training dataset
* Shared with team members for handoff
* Logged into your CRM or ticketing system via webhook

***

### 🌍 Phase 4 – External Live Chat Deployment

Once the agent achieves high internal accuracy (≥95%) and positive team feedback:

#### ✅ Enable Live Chat Skill

* Embed chat widget on internal or external site
* Set welcome message and branding
* Define escalation paths and operating hours

#### 🔁 Use Copilot for Dual Purposes

* Continue **internal Q\&A** and training
* **Monitor external chats in real time**
* Intervene or take over if needed

***

### ✅ Launch & Governance Checklist

| Area                                                               | Status |
| ------------------------------------------------------------------ | ------ |
| Documents transformed via Academy                                  | ✅      |
| Agent created in Launch Pad                                        | ✅      |
| Clear instructions defined                                         | ✅      |
| Packs uploaded and structured                                      | ✅      |
| Skills enabled (Copilot, Function, Notification, Scoring, Webhook) | ✅      |
| API connections tested                                             | ✅      |
| Copilot rollout complete                                           | ✅      |
| Scoring + summaries enabled                                        | ✅      |
| Live Chat deployment planned/tested                                | ✅      |

***

### 💡 Example Support Workflow

1. **Support agent** asks: "Where do I find the device log files?"
2. **AI Agent** replies with the exact SOP from your KB.
3. Agent marks the answer 👍 → added to training loop.
4. A conversation needs follow-up → Scoring Skill marks it.
5. **Notification** sent to the human support lead via SMS.
6. Conversation logged into **Jira** using the **Webhook Skill**.
7. Agent eventually supports **external customers** via Live Chat.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.raiaai.com/dev-program/agents/building-an-support-ai-agent.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.