Skip to main content

​
Prompt Guide for Specific Situations

Here you will find specific prompt guides and examples for common scenarios you might encounter when building a voice assistant with Famulor.

​
Pronouncing Phone Numbers

Use the Speak Slowly Feature

We use the “speak slowly” feature to add pauses between words. For more information, see the Speech Controllability Documentation. It is recommended to use a prompt as a guideline for the LLM. This ensures the assistant consistently responds with the correct format, even when customers verify the phone number.

​
Example for Pronouncing Phone Numbers

​
Spelling Email Addresses

Pronounce Email Addresses Correctly

Email addresses must be spelled out to avoid confusion. Here is the correct approach:

​
Example for Spelling Email Addresses

​
Pronouncing Websites

Pronounce Website URLs Correctly

Website URLs must be pronounced clearly and understandably. Here is the correct approach:

​
Example for Pronouncing Websites

​
Pronouncing Times

Pronounce Times Correctly

Times must be pronounced in an understandable format. Here are the guidelines:

​
Example for Pronouncing Times

​
Putting Customers On Hold / No Response Needed

Putting Customers on Hold

Sometimes you need to put customers on hold or no response is required. Famulor has a special function for this:

​
For Non-Reasoning Models

​
For Reasoning Models

Reasoning models like gpt-5 or gpt-5.1 do not support this feature. You must craft prompts differently to achieve this behavior.

​
Normalizing Text for Speech

Normalize Text for Speech

Normalize certain parts of the text (numbers, currency, dates, etc.) into their spoken form for more consistent speech synthesis. Sometimes TTS models misread non-normalized text.

​
Example for Text Normalization

This feature adds a small latency (~100ms) to the overall process.

​
Language Support

Supported Languages

Currently, text normalization is supported for the following languages:
  • German - full support
  • English - full support
  • Spanish - full support
  • French - full support
Other languages result in a no-op (text is not modified) for this step.

Language Settings

When you select a language other than multilingual, that language code is used to normalize the text (e.g., 1 is normalized to “eins” when using German). When you select multilingual, the language is detected automatically based on the generated text and normalized accordingly.

​
Prompt Engineering Guide

Best Practices for Effective Prompts

Best practices for writing effective prompts that create reliable and natural-sounding AI phone assistants.

​
Introduction

Prompt engineering is the foundation for creating effective AI phone assistants. A well-designed prompt determines how your assistant interprets situations, responds to users, and handles edge cases. This guide provides proven strategies for writing prompts your assistant can reliably follow.
This guide focuses on general prompt engineering principles. For agent-specific implementation:
  • Single/Multi Prompt Agents: Apply these principles directly in your prompts
  • Conversation Flow Agents: Use these principles within individual node instructions

​
Best Practice 1: Use Structured Prompts

Break large prompts into focused sections to improve organization and LLM understanding. This structured approach offers several advantages:
  • Reusability: Sections can be adapted across different agents
  • Maintainability: Easily update specific behaviors without affecting others
  • Clarity: LLMs process structured information more accurately
## Identity

You are a friendly AI assistant for [Company Name].

Your role is [specific purpose].

You have expertise in [relevant areas].

## Style Guidelines

Be concise: Keep answers under 2 sentences unless explaining complex topics.

Be conversational: Use natural language, contractions, and acknowledge what the caller says.

Be empathetic: Show understanding for the caller's situation.

## Response Guidelines

Return data in spoken form: say "fifteenth January" not "15.01".

Ask one question at a time: avoid overwhelming the caller with multiple questions.

Confirm understanding: paraphrase key information back to the caller.

## Task Instructions

[Specific steps the assistant should follow]

## Objection Handling

If the caller says they are not interested: "I understand. Is there something specific..."

If the caller is frustrated: "I understand your frustration, let me help solve this..."

​
Best Practice 2: Use Conversation Flow for Complex Tasks

If your assistant needs to handle complex logic or multiple tools, use Conversation Flow Agents instead of managing everything in a single prompt.

​
When to switch to Conversation Flow:

  • Multiple decision branches: More than 3-4 conditional paths
  • Tool coordination: Using 5+ different functions/tools
  • State management: Tracking multiple variables over conversation
  • Reliability concerns: Single prompt exhibits inconsistent behavior

​
Benefits of Conversation Flow:

  • Each node focuses on one specific task
  • Deterministic tool call and transitions
  • Easier to debug and optimize individual steps
  • More predictable assistant behavior

​
Best Practice 3: Explicit Tool Call Instructions

This section applies only to Single/Multi Prompt Agents. Conversation Flow Agents handle function calls deterministically via their node configuration.

​
The Challenge

LLMs often struggle to determine when to call tools based only on tool descriptions. Without explicit instructions, agents may:
  • Call tools at inappropriate times
  • Fail to call tools when needed
  • Use the wrong tool for a situation

​
Solution: Define Clear Triggers

Always provide precise conditions for tool usage in your prompts. Refer to tools by their exact names.
Example: Customer Service Agent
## Tool Usage Instructions

1. First, gather information about the customer's issue.

2. Determine the type of request:

   - If the customer mentions "refund" or "money back":

     → Call the function `transfer_to_support` immediately

   - If the customer needs order status:

     → Call the function `check_order_status` with order_id

   - If the customer wants to change their order:

     → First call `check_order_status`

     → Then switch to change status

3. After retrieving information:

   - Always summarize what you found

   - Ask if they need additional help

   - If yes, determine the next appropriate action

​
Best Practices for Tool Instructions

  1. Use trigger words: List specific words/phrases that should trigger tool calls
  2. Define sequences: Specify when tools should be called in order
  3. Set boundaries: Clarify when certain tools SHOULD NOT be called
  4. Provide context: Explain why each tool is called

​
Practical Examples for Famulor

Famulor-Specific Adjustments

Here are some Famulor-specific customizations:

​
Example of a Complete Prompt

​
Testing and Optimizing

Test Your Prompts

  • Conduct audio tests - Listen to the pronunciation
  • Test various scenarios - Cover all specific situations
  • Collect customer feedback - Ask about understandability
  • Iteratively improve - Adjust based on results

​
Help and Support

Our support team can help you customize your prompts for specific situations.

Contact Support

Need help customizing your prompts? Contact our support team.

​
Next Steps

Further Resources