chatbot-builder

Chatbot Builder

Overview

Designs complete chatbot architectures with conversation flows, intent mapping, response templates, and escalation rules.

Workflow

Step 1: Define the Bot

1. Purpose: Customer support, lead qualification, FAQ, onboarding, or sales?
2. Platform: Website widget, WhatsApp, Messenger, Slack, or custom?
3. Persona: What's the bot's personality?
4. Scope: What topics should it handle? What should it NOT handle?
5. Escalation: When should it hand off to a human?

Step 2: Map Intents

Intent	Example Phrases	Priority
Greeting	"hi", "hello", "hey there"	High
Pricing	"how much", "pricing", "cost"	High
Support	"help", "problem", "not working"	High
[Custom]	[phrases]	[priority]

Step 3: Design Conversation Flows

USER: [trigger phrase]
  ↓
BOT: [response + follow-up question]
  ↓
USER: [option A] → BOT: [response A] → [next step]
USER: [option B] → BOT: [response B] → [next step]
USER: [unclear] → BOT: [clarification] → [retry]
  ↓
[3 unclear attempts] → ESCALATE TO HUMAN

Step 4: Write Response Templates

For each intent:

• Initial response: First message
• Follow-up questions: Gather needed info
• Resolution: Answer or action
• Fallback: When bot can't help
• Handoff message: Transition to human

Output Format

# Chatbot Design: [Bot Name]

## Bot Profile
- **Name**: [bot name]
- **Personality**: [description]
- **Platform**: [where it lives]
- **Supported languages**: [languages]

## Intent Map
[Table of all intents with examples and responses]

## Conversation Flows

### Flow 1: [Name]
[Visual flow with decision trees]

### Flow 2: [Name]
[Visual flow]

## Response Templates
[All response messages organized by intent]

## Escalation Rules
- After [X] failed attempts → human
- Keywords [list] → immediate human
- Sentiment negative → human
- Off-hours → ticket + follow-up

## Knowledge Base
[FAQ answers the bot should know]

## Implementation Notes
- Platform: [recommended tool]
- AI model: [if using LLM]
- Integration: [APIs needed]

Quality Checklist

• [ ] All common intents mapped
• [ ] Fallback responses for unrecognized input
• [ ] Escalation rules clear
• [ ] Persona consistent across responses
• [ ] Tested with 10+ conversation paths