Wevolt Support Command Centre

An AI-powered support automation platform that ingests tickets from Zoho Desk, drafts replies using GPT, routes them through a human review workflow, and sends approved responses directly to customers.

Overview

The Support Command Centre sits between Zoho Desk and your customers. When a new ticket arrives in Zoho, a webhook fires to this system. The AI pipeline takes over — detecting spam, categorising the issue, assessing priority, retrieving relevant knowledge, and drafting a reply. A human reviewer then approves, edits, escalates, or requests a revision before anything is sent to the customer.

Key capabilities

  • Automatic spam detection with AI + keyword rules
  • Category and priority assignment for every ticket
  • Knowledge base and similar-ticket context injected into every draft
  • One-click approve / escalate / revise via email — no login required
  • Approved replies sent directly to Zoho Desk and emailed to the customer
  • Auto-close rules for inactive or resolved tickets (configurable)
  • Full audit trail for every action taken

Ticket Pipeline

Every ticket goes through a sequential processing pipeline managed by background workers:

1

Ingest

Zoho Desk sends a webhook when a ticket is created or updated. The system fetches the ticket thread from Zoho and stores it locally.

2

Spam Detection

GPT-4o mini checks the subject and message for spam. Keyword patterns are checked first (instant). Spam tickets are silently discarded — no draft is generated.

3

Categorisation

GPT-4o mini assigns one of 15 categories: Billing, RFID/Cards, App Issues, Charger Offline, Charging Failed, OCPP Connectivity, Load Management, Metering/MQTT, Commissioning, OCPI/Roaming, Whitelabel Support, Complaint, Feature Request, General Inquiry, Spam.

4

Priority Assessment

GPT-4o mini assigns urgent high medium low priority based on the category, subject, and message content.

5

Knowledge Retrieval

The system searches the knowledge base (KB articles, past resolved tickets) for relevant content to include in the draft context.

6

Draft Generation

GPT-5 mini generates a reply draft using the ticket thread, knowledge context, and the company agent briefing configured in Settings.

7

Review Notification

A review email is sent to the configured reviewer with the draft, ticket details, and one-click action links (Approve / Escalate / Revise).

8

Send Reply

Once approved, the reply is posted to Zoho Desk via sendReply and emailed to the customer from support@wevolt.zohodesk.com.au.

Review Queue

The Review Queue (/queue) shows all AI-generated draft replies awaiting human review. Tickets are listed with their summary, AI model used, confidence score, draft version, and current status.

Status meanings

pendingAwaiting reviewer action
approvedApproved — reply queued for sending
repliedReply sent to customer via Zoho
closedTicket closed — no further action needed
escalatedEscalated to L2 support team
revisedRevision requested — AI is redrafting
editedManually edited by reviewer
spamDetected as spam — no reply sent

Click View → on any row to open the review detail page where you can read the full draft and take action.

Review Actions

On the review detail page you have five actions:

Approve

Sends the draft reply exactly as shown to the customer via Zoho Desk. The ticket status moves to replied.

Edit & Approve

Opens an inline editor so you can modify the draft text before sending. Saves the edited version to Zoho.

Request Revision

Provide instructions (e.g. "be shorter", "add RFID steps") and the AI will regenerate the draft. The new draft is requeued for review. You can revise multiple times.

Escalate to L2

Marks the ticket as escalated and notifies the L2 escalation email address. No reply is sent to the customer automatically.

Mark as Spam

Manually marks the ticket as spam. No reply is sent. Use this if the AI missed a spam ticket.

Email Actions (No Login Required)

When a review is ready, the system emails the reviewer with the draft and three one-click action links. These links are signed with a JWT token (valid for 72 hours) and do not require the reviewer to log in.

Approve

One click — immediately sends the draft to the customer.

Escalate

One click — escalates the ticket to L2 without sending a customer reply.

Request Revision

Opens a form in the browser where you can type revision instructions. The AI re-drafts and re-sends the review email with the updated draft.

Email action links expire after 72 hours. If a link has expired, open the Review Queue in the dashboard and take action there instead.

Close approval emails are also sent when the auto-close monitor flags a ticket. These have Confirm Close and Keep Open links.

Knowledge Base

The Knowledge Base (/kb) shows all indexed sources used by the AI when generating draft replies. Two types of sources are supported:

Zoho KB Articles

Synced daily (2 AM UTC) from your Zoho Desk knowledge base. Articles are chunked, embedded, and stored in a vector database for semantic search.

Custom Sources

You can add custom text or document sources directly from the KB page. These are chunked and indexed immediately.

During draft generation, the top 8 most semantically relevant KB chunks are retrieved and injected into the AI prompt alongside the ticket thread.

AI Chat

The AI Chat (/chat) lets support agents ask questions directly to the AI assistant, with full access to the knowledge base for context. Use it to:

  • Look up how to handle specific charger error codes
  • Draft manual replies for tickets not in the system
  • Understand policies (T&C, privacy, warranty)
  • Ask anything about the Wevolt platform

Chat history is per-session and is not persisted after you navigate away.

Audit Log

The Audit Log (/activity) records every action taken by the AI and by reviewers. Each entry shows the ticket ID, event type, timestamp, and a JSON payload with full details.

Event types logged

ticket_ingestedspam_detectioncategorizationpriority_assignmentknowledge_retrieveddraft_generateddraft_revisedreview_approved_via_emailreview_escalated_via_emailrevision_requested_via_emailreply_sentticket_closed_via_emailclose_dismissed

Settings

Access settings at /settings. The following are configurable:

Agent Briefing

A system prompt injected into every draft generation call. Use this to describe your company, tone of voice, what not to say, product specifics, and sign-off instructions. The AI will follow these instructions on every reply.

T&C & Privacy Policy

Summaries of your Terms and Conditions and Privacy Policy. The AI references these when answering policy-related questions. You can provide either a URL or paste the full text.

Email Routing

Configures where review notification emails are sent. Routes include: review (new draft ready), escalation (L2 escalation), and close_approval (auto-close confirmation).

Department Routing

Configures which Zoho Desk department tickets are moved to after triage. Default routes:

  • L1 Customer Support — Billing, App Issues, RFID/Cards, Charging Failed, General Inquiry, Complaint
  • L2 Tech Support — Charger Offline, OCPP Connectivity, Load Management, Metering/MQTT, Commissioning, OCPI/Roaming, Whitelabel Support
  • Sales — Feature Request
  • Spam Department — Spam-detected tickets

Each route is configurable — you can set the Zoho Department ID, update which categories trigger it, and enable/disable routes individually.

Auto-Close Rules

Two rules can be enabled independently:

  • No response after N days — close tickets that have been in "replied" status for N days with no customer follow-up.
  • Customer confirmed resolved — close tickets where the AI detected the customer said the issue is fixed.

Require human approval (default ON) sends a confirmation email before closing. With approval off, tickets are closed automatically.

AI Models

GPT-4o mini — Triage tasks

Used for: spam detection, categorisation, priority assessment, moderation checks. Fast and cost-efficient for classification tasks.

GPT-5 mini — Draft generation

Used for: writing customer reply drafts, processing revision instructions, escalation summaries, AI chat responses. Higher quality output for customer-facing content.

text-embedding-ada-002 — Vector search

Used for: KB article similarity search, finding similar past tickets. Powers semantic retrieval for draft context.

All models are served via Azure AI Foundry (Azure OpenAI). No data is sent to OpenAI directly.

Current Limitations

Known limitations to be aware of:

  • Single reviewer — the review email goes to one configured address per route. Multi-reviewer workflows are not yet supported.
  • Reply only to email tickets — replies are sent via the Zoho sendReply EMAIL channel. Tickets submitted via web form may not forward an email to the customer depending on Zoho's configuration.
  • No attachment handling — images or files attached to tickets are not read by the AI. The AI only sees text content.