RAG & Knowledge Base: Powering Your AI with Specific Information

RAG (Retrieval-Augmented Generation) is what makes your AI agents truly powerful. Instead of relying on general knowledge, RAG allows your agents to access and reference specific documents, policies, and information that matter to your work.

What is RAG?

RAG stands for Retrieval-Augmented Generation. It's a technique that combines:

Retrieval: Finding relevant information from your documents
Augmentation: Adding that information to the AI's context
Generation: Creating responses based on your specific information

Why RAG Matters

Without RAG, AI agents rely only on their general training data. With RAG, they can:

Answer questions about your specific services and policies
Reference current information and procedures
Provide accurate, up-to-date responses
Access information that wasn't available when the AI was trained

How RAG Works in Signpost AI

The Process

User asks a question → "What services are available for refugees?"
Search Worker finds relevant documents → Searches your refugee services collection
AI receives context → Gets information about available services
AI generates response → Creates answer based on your specific information

Example: Refugee Services Assistant

User: "What housing assistance is available for asylum seekers?"

1. Search Worker searches your "Refugee Services" collection
2. Finds documents about housing programs, eligibility, and application processes
3. AI receives this context: "Housing assistance includes temporary shelter,
   rental assistance, and case management. Eligibility requires asylum application..."
4. AI responds: "Based on our current programs, housing assistance for asylum
   seekers includes temporary shelter, rental assistance, and case management
   services. To be eligible, you must have filed an asylum application..."

Creating Collections for Your Knowledge Base

What is a Collection?

A collection is a group of related documents that your AI agent can search through. Think of it as a specialized library for a specific topic.

Example Collections for Refugee Services

1. Legal Services Collection

Documents to include:

Asylum application procedures
Legal rights and responsibilities
Immigration court information
Pro bono legal service providers
Document requirements and deadlines

2. Housing & Basic Needs Collection

Documents to include:

Housing assistance programs
Emergency shelter locations
Food assistance programs
Transportation services
Utility assistance programs

3. Healthcare Services Collection

Documents to include:

Medical care access
Mental health services
Vaccination programs
Health insurance information
Clinic locations and hours

4. Education & Employment Collection

Documents to include:

English language classes
Job training programs
School enrollment information
Employment rights
Professional certification programs

5. Community Resources Collection

Documents to include:

Community organizations
Cultural support groups
Religious services
Social events and activities
Volunteer opportunities

How to Add Sources and Create Collections

Step 1: Plan Your Collections

Start with your most common questions:

What services do people ask about most?
What information changes frequently?
What documents do your staff reference often?

Example Planning:

Collection: "Refugee Legal Services"
Purpose: Help with legal questions and procedures
Common Questions:
- How do I apply for asylum?
- What documents do I need?
- How long does the process take?
- Can I get free legal help?

Step 2: Gather Your Documents

Types of documents to include:

Policies and Procedures: Official guidelines and processes
FAQs: Frequently asked questions and answers
Forms and Applications: Official documents and requirements
Contact Information: Phone numbers, addresses, office hours
Eligibility Criteria: Who qualifies for different services
Timelines and Deadlines: Important dates and timeframes

Document Formats Supported:

PDF files
Word documents (.docx)
Text files (.txt)
Web pages (URLs)
Spreadsheets (.csv, .xlsx)

Step 3: Create Your Collection

Go to Knowledge → Collections
Click "Create New Collection"
Name your collection (e.g., "Refugee Legal Services")
Add a description (e.g., "Legal services, asylum procedures, and immigration support")
Upload your documents

Step 4: Organize Your Information

Best Practices:

Use clear file names: "Asylum_Application_Process_2024.pdf"
Group related documents: All housing info in one collection
Keep information current: Update documents regularly
Include multiple languages: If serving diverse communities

Real-World Example: Refugee Services Knowledge Base

Collection Structure

📁 Refugee Services Knowledge Base
├── 📁 Legal Services
│   ├── Asylum_Application_Guide.pdf
│   ├── Legal_Rights_Handbook.pdf
│   ├── Pro_Bono_Lawyers_List.pdf
│   └── Court_Procedures.pdf
├── 📁 Housing & Basic Needs
│   ├── Housing_Assistance_Programs.pdf
│   ├── Emergency_Shelter_Locations.pdf
│   ├── Food_Assistance_Guide.pdf
│   └── Transportation_Services.pdf
├── 📁 Healthcare
│   ├── Medical_Services_Guide.pdf
│   ├── Mental_Health_Resources.pdf
│   ├── Vaccination_Schedule.pdf
│   └── Clinic_Locations.pdf
└── 📁 Education & Employment
    ├── English_Classes_Schedule.pdf
    ├── Job_Training_Programs.pdf
    ├── School_Enrollment_Guide.pdf
    └── Employment_Rights.pdf

Sample Documents for Refugee Services

Asylum Application Guide

Title: "Asylum Application Process - 2024"
Content includes:
- Step-by-step application process
- Required documents checklist
- Filing deadlines and timelines
- Common mistakes to avoid
- Where to get help

Housing Assistance Programs

Title: "Housing Assistance for Refugees and Asylees"
Content includes:
- Available housing programs
- Eligibility requirements
- Application procedures
- Required documentation
- Contact information for housing agencies

Using RAG in Your AI Agents

Connecting Collections to Your Agent

Add a Search Worker to your agent flow
Configure the Search Worker:
- Select your collection (e.g., "Refugee Legal Services")
- Set search parameters
- Connect to your AI Worker

Example Agent Flow

User Input → Search Worker → AI Worker → Response
                ↓
         Searches "Refugee Legal Services" collection
                ↓
         Finds relevant documents about asylum process
                ↓
         AI generates response using found information

Search Worker Configuration

Search Parameters:

Collection: Which collection to search
Search Query: What to look for (usually from user input)
Number of Results: How many documents to retrieve
Similarity Threshold: How closely documents must match

Example Configuration:

Collection: "Refugee Legal Services"
Search Query: "asylum application process"
Number of Results: 3
Similarity Threshold: 0.7

Best Practices for RAG Implementation

1. Organize Information Logically

Good Organization:

📁 Legal Services
├── 📄 Asylum Process
├── 📄 Legal Rights
└── 📄 Court Information

📁 Housing Services
├── 📄 Emergency Housing
├── 📄 Long-term Housing
└── 📄 Housing Rights

Poor Organization:

📁 All Documents
├── 📄 Document1.pdf
├── 📄 Document2.pdf
└── 📄 Document3.pdf

2. Keep Information Current

Regular Updates:

Review documents monthly
Update contact information
Remove outdated information
Add new services and programs

3. Use Clear, Searchable Content

Good Document Content:

"To apply for asylum, you must file Form I-589 within one year of arriving in the United States.
Required documents include: passport, birth certificate, evidence of persecution, and supporting affidavits."

Poor Document Content:

"Apply for asylum using the form. Include documents."

4. Test Your Knowledge Base

Regular Testing:

Ask common questions
Verify responses are accurate
Check that relevant documents are found
Update based on user feedback

Advanced RAG Features

Multi-Collection Search

Search across multiple collections for comprehensive answers:

User: "What services are available for a family with children?"

Search Worker searches:
- Legal Services collection
- Housing & Basic Needs collection
- Healthcare collection
- Education & Employment collection

AI receives information from all collections and provides comprehensive response

Context-Aware Search

Use conversation history to improve search results:

User: "I need help with housing"
AI: "I can help you with housing assistance. What type of housing do you need?"

User: "I have children"
AI: "For families with children, we have several housing options including family shelters,
     transitional housing, and rental assistance programs. Let me search for specific details..."

Language-Specific Collections

Create collections in different languages:

📁 Refugee Services (English)
📁 Refugee Services (Spanish)
📁 Refugee Services (Arabic)
📁 Refugee Services (French)

Troubleshooting RAG Issues

Common Problems and Solutions

1. AI Not Finding Relevant Information

Problem: Search worker returns irrelevant documents Solutions:

Improve document organization
Use more specific collection names
Add better document titles and descriptions
Adjust similarity threshold

2. AI Ignoring Retrieved Information

Problem: AI generates responses without using found documents Solutions:

Improve your prompt to emphasize using provided information
Add instructions: "Only use information from the provided documents"
Check that Search Worker is properly connected to AI Worker

3. Outdated Information

Problem: AI provides outdated information Solutions:

Regular document updates
Version control for important documents
Clear dates on all documents
Archive old versions

4. Poor Search Results

Problem: Search worker doesn't find the right documents Solutions:

Improve document content quality
Use more descriptive file names
Add metadata and tags
Test different search queries

Getting Started with RAG

Week 1: Plan and Organize

Identify your most common questions
Gather existing documents and information
Plan your collection structure
Create your first collection

Week 2: Build Your Knowledge Base

Upload and organize documents
Create multiple collections
Test search functionality
Refine organization

Week 3: Integrate with Your Agent

Add Search Worker to your agent
Configure search parameters
Test with real questions
Adjust and improve

Week 4: Optimize and Expand

Monitor search performance
Add more documents and collections
Train staff on new capabilities
Plan for ongoing maintenance

Ready to build your knowledge base? Start with your most common questions and build from there!

On this page