Non-Developer Guide to Signpost AI

This guide is designed for non-technical users who want to understand how Signpost AI works, why it's valuable, and how to use it effectively. No coding knowledge required!

What is Signpost AI and Why Should You Care?

Signpost AI is a visual AI platform that lets you build intelligent applications without writing code. Think of it as having a team of AI experts working for you, but without the complexity of traditional software development.

Why This Matters for Non-Developers

No Coding Required: Build powerful AI applications using a visual interface
Rapid Results: Create working solutions in hours, not months
Cost Effective: Avoid expensive developer hiring and long development cycles
Flexible: Adapt and change your applications as your needs evolve

How Everything Works Together

The Building Block Approach

Signpost AI uses a "building block" system where each component has a specific job:

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Input Block   │───▶│  AI Processing  │───▶│  Output Block   │
│   (User Query)  │    │   (Brain Work)  │    │   (Response)    │
└─────────────────┘    └─────────────────┘    └─────────────────┘

The Flow of Information

User sends a message → Input block receives it
AI processes the message → Uses your instructions and knowledge base
AI generates a response → Creates a helpful answer
Response is sent back → User receives the answer

Real-World Example: Community Health Assistant

User: "What vaccines are available for children?"

Input Block: Receives the question
AI Processing:
  - Checks your health information database
  - Uses your instructions about being helpful and accurate
  - Generates a response based on available vaccines
Output Block: Sends back the vaccine information

The Power of Prompting

What is a Prompt?

A prompt is simply the instructions you give to your AI. It's like writing a job description for a very smart assistant.

The LLM Agent: The Core of Your AI System

Think of the LLM Agent as the "brain" of your Signpost AI system. It's not just another component - it's the central decision-maker that makes everything work together intelligently.

Why We Built Around the LLM Agent

The Problem We Solved: Traditional AI systems often work like a rigid assembly line - each step happens in isolation, with no understanding of context or the bigger picture. This leads to:

Fragmented responses that don't make sense together
Lost context between different parts of the conversation
Inefficient processing that doesn't learn from the interaction
Poor user experience with robotic, disconnected responses

Our Solution: We made the LLM Agent the central orchestrator that:

Understands the full context of every conversation
Makes intelligent decisions about what information to use
Coordinates all other components (search, document generation, messaging)
Maintains conversation memory and user intent
Provides natural, coherent responses

How the LLM Agent Works as Your Core

┌─────────────────────────────────────────────────────────────┐
│                    LLM Agent (The Brain)                    │
│  ┌─────────────────┐  ┌─────────────────┐  ┌──────────────┐ │
│  │   Understands   │  │   Coordinates   │  │   Generates  │ │
│  │   User Intent   │  │   All Workers   │  │   Responses  │ │
│  └─────────────────┘  └─────────────────┘  └──────────────┘ │
└─────────────────────────────────────────────────────────────┘
         │                    │                    │
         ▼                    ▼                    ▼
┌──────────────┐    ┌──────────────┐    ┌──────────────┐
│   Search     │    │  Document    │    │   Message    │
│   Worker     │    │  Generator   │    │   Worker     │
│ (Finds Info) │    │ (Creates)    │    │ (Sends)      │
└──────────────┘    └──────────────┘    └──────────────┘

The Context and Efficiency Advantage

Context Preservation:

Traditional Approach: Each component works independently, losing context
Our Approach: LLM Agent maintains full conversation context and user intent

Example - Traditional vs. Our Approach:

Traditional (Fragmented):

User: "I need help with my child's vaccination"
System: [Search finds vaccine info] → [Separate component formats response] → [Another component sends it]
Result: Generic vaccine information without considering the child's age, location, or specific needs

Our LLM Agent Approach:

User: "I need help with my child's vaccination"
LLM Agent: "This parent needs specific vaccination guidance. Let me:
1. Search for age-appropriate vaccines
2. Check what's available in their area
3. Consider their previous questions about health
4. Generate a personalized, helpful response"
Result: Tailored vaccination guidance that considers the full context

Efficiency Benefits:

Smarter Resource Usage:
- Only searches for relevant information
- Avoids redundant processing
- Uses the most appropriate tools for each situation
Better User Experience:
- Maintains conversation flow
- Remembers previous interactions
- Provides coherent, contextual responses
Reduced Costs:
- More efficient use of AI processing
- Fewer unnecessary API calls
- Better resource allocation

Real-World Example: Community Health Assistant

Scenario: A parent asks about their 2-year-old's vaccination schedule

LLM Agent's Intelligent Process:

Context Understanding: "This is a parent concerned about their toddler's health"
Smart Coordination:
- Searches for age-specific vaccine information
- Checks local clinic availability
- Considers cultural and language preferences
- Prepares a warm, informative response
Efficient Response:
- Provides exactly what the parent needs
- Suggests next steps (appointment booking, etc.)
- Maintains helpful, caring tone throughout

Result: The parent gets a comprehensive, personalized response that feels like talking to a knowledgeable, caring health worker.

Why This Matters for Non-Developers

You Don't Need to Worry About:

Complex technical integration between components
Managing conversation state across different systems
Ensuring consistent tone and personality
Coordinating multiple AI services

You Just Need to:

Write clear instructions for your LLM Agent
Provide good information in your knowledge base
Test and refine your prompts

The LLM Agent Handles:

All the complex coordination behind the scenes
Maintaining conversation context
Making intelligent decisions about what tools to use
Ensuring consistent, helpful responses

The Prompting Advantage

Because the LLM Agent is the core, your prompts become incredibly powerful:

Simple Prompt Example:

"You are a community health worker. Help families with health information and connect them to services."

The LLM Agent Automatically:

Understands when to search for information
Knows when to generate documents
Decides when to escalate to human help
Maintains appropriate tone and context
Coordinates all the technical details

This is why prompting is so powerful in Signpost AI - you're not just giving instructions to one component, you're programming the entire intelligent system that coordinates everything else.

Why Prompting is So Powerful

Good Prompt Example:

"You are a compassionate community health worker. Your job is to help people
understand health information and access available services. Always be accurate,
helpful, and culturally sensitive. If you don't know something, say so and
offer to connect them with a human health worker."

Poor Prompt Example:

"Help with health questions"

The Difference Good Prompting Makes

Aspect	Poor Prompt	Good Prompt
Tone	Generic, robotic	Warm, professional
Accuracy	May make up information	Admits when unsure
Cultural Sensitivity	One-size-fits-all	Adapts to context
Safety	No boundaries	Clear escalation paths

Prompting Best Practices

1. Be Specific About the Role

❌ "Help people"
✅ "You are a community health educator helping families understand vaccination schedules"

2. Set Clear Boundaries

❌ "Answer any question"
✅ "You can help with health information and service referrals. For medical emergencies, transfer to a human health worker."

3. Include Examples

"Here are examples of good responses:
- 'Based on our clinic information, we offer measles and polio vaccines for children aged 6 months to 5 years.'
- 'I don't have specific information about that medication. Let me connect you with our pharmacist.'"

4. Consider Cultural Context

"Always be respectful of cultural beliefs and practices. Ask clarifying questions if needed to provide appropriate guidance."

The Trial and Error Process: Mastering Prompting

Why Trial and Error is Essential

Prompting is more art than science. Even the most experienced AI users go through multiple iterations to get the best results. Think of it like teaching someone new - you need to try different approaches to see what works best.

The Iterative Process

Step 1: Start with a Basic Prompt

Example for a Community Health Assistant:

"You are a health assistant. Help people with health questions."

Test it: Ask questions like "What vaccines do I need?" and see how it responds.

Step 2: Identify Problems

Common Issues You Might See:

Too generic: "I can help with health questions" (not specific enough)
Too formal: "I am a healthcare professional..." (not approachable)
No boundaries: Gives medical advice it shouldn't
No personality: Sounds robotic and cold

Step 3: Refine and Test Again

Improved Version:

"You are a friendly community health assistant. Your role is to:
- Provide general health information from our clinic database
- Help people understand vaccination schedules
- Connect people to appropriate services
- NEVER give specific medical advice or diagnoses
- Always be warm, patient, and culturally sensitive

If someone asks for medical advice, say: 'I can provide general information, but for specific medical advice, please consult with a healthcare provider.'"

Step 4: Test with Real Scenarios

Test Questions to Try:

"What vaccines do children need?"
"I have a headache, what should I do?"
"Can you diagnose my rash?"
"What services do you offer for pregnant women?"
"I'm worried about my child's fever"

Common Prompting Patterns and Solutions

Pattern 1: AI is Too Vague

Problem: AI gives generic responses like "I can help with that"

Solution: Add specific instructions and examples

❌ "Help with health questions"
✅ "Provide specific information about our clinic's services. For example:
- Vaccination schedules: 'We offer measles vaccines for children 12-15 months'
- Service referrals: 'For prenatal care, contact our women's health clinic at 555-0123'
- General info: 'Our clinic is open Monday-Friday, 8 AM to 5 PM'"

Pattern 2: AI Makes Up Information

Problem: AI invents facts or services you don't offer

Solution: Set clear boundaries and reference requirements

❌ "Answer health questions"
✅ "Only provide information that exists in our clinic database. If you don't have specific information, say: 'I don't have that information in our database. Let me connect you with a staff member who can help.'"

Pattern 3: AI is Too Formal

Problem: Responses sound robotic and unfriendly

Solution: Add personality and tone instructions

❌ "You are a health assistant"
✅ "You are a warm, friendly community health worker who speaks in a caring, approachable way. Use simple language and show empathy. For example, instead of 'Vaccination is recommended,' say 'Getting vaccinated helps keep your family safe and healthy.'"

Pattern 4: AI Doesn't Handle Edge Cases

Problem: AI gets confused by unexpected questions

Solution: Provide fallback instructions

"Your main role is helping with health information. If someone asks about:
- Non-health topics: 'I'm here to help with health information. Is there something health-related I can assist you with?'
- Medical emergencies: 'If this is a medical emergency, please call emergency services immediately.'
- Personal problems: 'I can help with health information, but for personal counseling, I recommend speaking with a counselor or social worker.'"

Testing Strategies

1. Role-Playing Test

Have different people test your AI with various scenarios:

Friendly user: Asks simple questions politely
Confused user: Asks unclear or complex questions
Urgent user: Asks emergency-related questions
Cultural user: Asks questions from different cultural perspectives

2. Edge Case Testing

Try questions that might break your AI:

"What's the meaning of life?"
"Can you help me hack into someone's account?"
"I want to hurt myself"
"Tell me a joke"
"What's the weather like?"

3. Consistency Testing

Ask the same question multiple times to check for consistency:

Does it give the same answer?
Does it maintain the same tone?
Does it follow the same format?

Iteration Checklist

After each test, ask yourself:

Does the response match your organization's values?
Is the information accurate and helpful?
Is the tone appropriate for your audience?
Does it handle edge cases gracefully?
Is it cost-effective (not too long or complex)?
Does it provide clear next steps when needed?

When to Stop Iterating

Stop when you have:

Consistent quality: 80%+ of responses are good
User satisfaction: Test users are happy with responses
Cost efficiency: Responses are reasonably priced
Safety: No harmful or inappropriate responses
Scalability: Can handle expected volume

Remember: Perfect is the enemy of good. Start with a working prompt and improve it over time based on real usage.

Tools for Testing

1. Signpost AI Playground

Test your prompts in real-time
See exactly how the AI responds
Try different variations quickly

2. User Feedback

Collect feedback from real users
Identify common issues
Understand user needs better

3. Analytics

Track which questions get poor responses
Monitor user satisfaction
Identify areas for improvement

Pro Tips for Faster Iteration

1. Start with Templates

Use existing prompt templates and customize them for your needs.

2. Test in Small Batches

Don't test everything at once. Focus on one aspect at a time.

3. Document Your Changes

Keep track of what works and what doesn't for future reference.

4. Learn from Others

Look at successful AI implementations in similar fields.

5. Be Patient

Good prompting takes time. Don't expect perfection on the first try.

Tradeoffs When Working with Different Workers

AI Workers (The Brain)

Pros:

Most flexible and powerful
Can handle complex conversations
Understands context and nuance

Cons:

Can be expensive for high volume
May generate unexpected responses
Requires careful prompting

Best For: Complex conversations, creative tasks, nuanced responses

Search Workers (The Researcher)

Pros:

Very fast and reliable
Always uses your specific information
Cost-effective for high volume

Cons:

Limited to your knowledge base
Can't handle complex reasoning
May not find relevant information

Best For: Factual questions, information retrieval, high-volume queries

Document Generator Workers (The Writer)

Pros:

Creates structured, formatted content
Consistent output format
Good for reports and summaries

Cons:

Less flexible than AI workers
Requires predefined templates
May not handle unexpected requests well

Best For: Report generation, form filling, structured content

When to Use Each Type

Use Case	Recommended Worker	Why
Complex customer support	AI Worker	Handles nuanced conversations
FAQ lookup	Search Worker	Fast, reliable, cost-effective
Report generation	Document Generator	Consistent formatting
Creative content	AI Worker	Flexible and creative
High-volume queries	Search Worker	Fast and scalable

Deterministic vs Generative AI

What's the Difference?

Deterministic AI:

Always gives the same answer for the same input
Like a calculator: 2+2 always equals 4
Predictable and reliable
Limited to predefined rules

Generative AI:

Can create different responses for the same input
Like a creative writer: can tell the same story in many ways
Flexible and creative
Learns from patterns in data

Real-World Examples

Deterministic Example (Search Worker):

User: "What vaccines are available?"
Response: "Measles, Polio, DPT" (Always the same)

Generative Example (AI Worker):

User: "What vaccines are available?"
Response 1: "We offer measles, polio, and DPT vaccines for children."
Response 2: "Based on our clinic schedule, children can receive measles, polio, and DPT vaccinations."
Response 3: "The vaccines we provide include measles, polio, and DPT - all important for child health."

When to Choose Each

Choose Deterministic When:

You need consistent, predictable responses
Accuracy is more important than creativity
You're dealing with factual information
Cost is a major concern

Choose Generative When:

You need natural, conversational responses
Context and nuance matter
You want to handle unexpected questions
User experience is a priority

What Can Go Wrong (And How to Prevent It)

Common Issues and Solutions

1. AI Makes Up Information

Problem: AI generates false or misleading information Solution:

Provide accurate information in your knowledge base
Add instructions: "Only use information from the provided documents"
Set up fact-checking workflows

2. Inappropriate Responses

Problem: AI gives responses that don't match your organization's values Solution:

Write clear guidelines in your prompts
Include examples of appropriate responses
Set up content filtering

3. Cost Overruns

Problem: AI usage costs more than expected Solution:

Use Search workers for simple queries
Set usage limits and monitoring
Choose cost-effective models for basic tasks

4. Poor User Experience

Problem: Users get frustrated with AI responses Solution:

Test with real users early and often
Provide clear escalation paths to humans
Monitor and improve based on feedback

5. Cultural Insensitivity

Problem: AI responses don't consider cultural context Solution:

Include cultural sensitivity in your prompts
Test with diverse user groups
Provide training data that reflects your community

Prevention Strategies

1. Start Small

Begin with simple, well-defined use cases
Test thoroughly before expanding
Learn from each iteration

2. Monitor Continuously

Track response quality and user satisfaction
Monitor costs and usage patterns
Watch for unexpected behaviors

3. Have Human Oversight

Set up review processes for important decisions
Provide clear escalation paths
Keep humans in the loop for critical functions

4. Regular Updates

Update your knowledge base regularly
Refine prompts based on user feedback
Stay current with best practices

Why Signpost AI is Valuable

For Organizations

Cost Savings:

Reduce manual work by 50-80%
Scale operations without proportional staffing increases
24/7 availability without overtime costs

Improved Service:

Instant responses to common questions
Consistent information delivery
Reduced wait times for users

Better Data:

Track user interactions and needs
Identify common questions and issues
Make data-driven improvements

For Communities

Accessibility:

Information available in multiple languages
24/7 access to services and information
Reduced barriers to getting help

Consistency:

Everyone gets the same accurate information
No variation based on staff availability
Standardized service delivery

Scalability:

Handle increased demand without adding staff
Serve more people with existing resources
Respond quickly during emergencies

Telerivet: Delivering Your Agent to Users

What is Telerivet and Why Does It Matter?

Telerivet is the messaging platform that connects your Signpost AI agent to real users through SMS and WhatsApp. Think of it as the "delivery truck" that brings your AI's responses directly to people's phones.

How Telerivet Works with Signpost AI

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   User's Phone  │◄──►│   Telerivet     │◄──►│  Signpost AI    │
│   (SMS/WhatsApp)│    │   (Messaging)   │    │   (Your Agent)  │
└─────────────────┘    └─────────────────┘    └─────────────────┘

The Flow:

User sends a message via SMS or WhatsApp
Telerivet receives it and forwards to your Signpost AI agent
Your agent processes the message using its intelligence
Agent sends response back through Telerivet
User receives response on their phone

Why Telerivet is Essential for Humanitarian Work

1. Universal Access

SMS works everywhere - even in areas with poor internet
WhatsApp is ubiquitous - used by billions globally
No app installation required - users just use what they already have

2. Reliable Communication

Works offline - messages are delivered when connection returns
Handles high volume - can manage thousands of conversations
Multilingual support - works in any language

3. Cost Effective

Low-cost messaging - much cheaper than building custom apps
No infrastructure needed - Telerivet handles the complexity
Scalable pricing - pay only for what you use

Real-World Example: Refugee Support Hotline

Scenario: A humanitarian organization provides support to refugees in a camp

How Telerivet + Signpost AI Works:

Refugee: "I need help finding medical care for my child"

Telerivet → Signpost AI: "User needs medical care information"

Signpost AI Agent:
- Searches knowledge base for local medical services
- Considers the refugee's location and needs
- Generates helpful response in appropriate language

Signpost AI → Telerivet: "Here are the medical clinics near your camp..."

Telerivet → Refugee: "Here are the medical clinics near your camp..."

Result: The refugee gets immediate, helpful information without needing internet, apps, or speaking to a human operator.

Setting Up Telerivet with Your Agent

What You Need to Configure:

Telerivet Account
- Sign up for Telerivet service
- Get your API credentials
- Set up your project
Integration Setup
- Connect Telerivet to your Signpost AI agent
- Configure message handling
- Set up response formatting
Testing
- Test with your own phone number
- Verify message delivery
- Check response quality