Chatbot - Building Vector Store

Overview

The Vector Store is a semantic search index that enables RAG (Retrieval-Augmented Generation). Without it, your chatbot uses only general AI knowledge. With it, your chatbot can find and reference your specific documents to provide accurate, context-aware answers.

What is a Vector Store?

Vector Store = Semantic Understanding

• Converts your documents into numerical representations (embeddings)
• Understands meaning and context, not just keywords
• Enables intelligent document retrieval
• Powers RAG-enhanced responses

Analogy: Think of it like an intelligent librarian who has read all your documents and can instantly find relevant information when asked questions, even if you don't use exact words from the documents.

Prerequisites

Before building a vector store:

1. Documents Must Be Completed

• All knowledge base items show status: "Completed"
• No items should be "Processing" or "Failed"
• Check the Knowledge Base Items table

2. Have Documents Uploaded

• At least one completed knowledge base item
• More documents = better knowledge base
• Quality documents work better than many poor ones

3. Chatbot Must Be Configured

• You've designed your chatbot (Step 1)
• System is ready for vector store

Building the Vector Store

Step-by-Step Process

1. Verify Knowledge Base Status

• Go to your Chatbot project page
• Scroll to "Knowledge Base Items" section
• Verify all items show "Completed"
• If any show "Processing", wait for them to complete
• If any show "Failed", delete and re-upload

2. Locate Vector Store Section

• Find "Vector Store" section on project page
• Shows current status
• Typically shows "Unavailable" if not built yet

3. Click "Build Vector Store"

• Button located in workflow or vector store section
• Clicking starts the build process
• Confirmation may appear

4. Wait for Processing

• Status changes to "Processing"
• Overlay may appear showing progress
• Can take 5-30 minutes depending on:
• Number of documents
• Total document size
• Server load

5. Monitor Progress

• Status updates automatically
• Page shows "Building..." or similar
• Don't close browser (optional, but recommended)

6. Completion

• Status changes to "Completed"
• Vector store is ready
• Can now use RAG features

What Happens During Building

1. Document Analysis

• System reads all your knowledge base items
• Extracts text content
• Identifies structure and sections

2. Chunking

• Documents split into smaller chunks
• Each chunk is a manageable size
• Overlapping chunks for better context

3. Embedding Generation

• Each chunk converted to vector embedding
• Numerical representation of meaning
• Uses AI models for semantic understanding

4. Index Creation

• Embeddings stored in vector database
• Indexed for fast retrieval
• Organized for efficient searching

5. Completion

• Vector store ready for queries
• RAG functionality enabled
• Can retrieve relevant chunks

Understanding Status Indicators

Vector Store Statuses

Unavailable:

• No vector store exists
• Need to build first
• RAG features not available

Processing:

• Building in progress
• Do not interrupt
• Wait for completion

Updating:

• Updating existing vector store
• Usually after adding/removing documents
• Faster than initial build

Completed:

• Ready to use
• RAG features enabled
• Can query documents

When to Rebuild Vector Store

Automatic Updates

In some cases, vector store updates automatically:

• After significant changes
• When documents added/removed
• System-triggered updates

Manual Rebuild

Rebuild When:

• Added many new documents
• Removed important documents
• Want to refresh embeddings
• Noticing poor RAG performance

How to Rebuild: 1. Click "Recreate Vector Store" or "Rebuild" 2. Confirm action 3. Wait for completion 4. Typically faster than initial build

Rebuild Process:

• May show "Updating" instead of "Processing"
• Preserves existing structure where possible
• Updates with new content

Processing Time

Factors Affecting Build Time

Document Count:

• Few documents (< 10): 5-10 minutes
• Medium (10-50): 10-20 minutes
• Many documents (50+): 20-30+ minutes

Document Size:

• Small documents: Faster processing
• Large documents: Slower processing
• Total content volume matters most

Server Load:

• Busy times: May take longer
• Off-peak: Usually faster
• Professional accounts may have priority

Estimated Times

| Documents | Estimated Time | |-----------|----------------| | 1-5 | 5-10 minutes | | 6-20 | 10-15 minutes | | 21-50 | 15-25 minutes | | 51-100 | 25-35 minutes | | 100+ | 30-60+ minutes |

Actual times may vary

Troubleshooting

Vector Store Won't Build

Problem: Documents Not Completed

• Solution: Wait for all items to show "Completed"
• Check for failed items
• Delete and re-upload failed items

Problem: Build Stuck in Processing

• Solution: Wait 30-45 minutes
• Refresh the page
• Contact support if stuck > 1 hour

Problem: Build Failed

• Solution: Check error message
• Verify documents are valid
• Try rebuilding
• Contact support if persists

Poor RAG Performance

Problem: Irrelevant Results

• Solution: Rebuild vector store
• Ensure documents are relevant
• Add more quality documents
• Check document quality

Problem: Missing Information

• Solution: Add more documents
• Rebuild vector store
• Ensure documents contain needed info

Using Vector Store

Once Built

RAG Features Enabled:

• Chatbot can query your documents
• Retrieves relevant information
• Combines with AI for answers
• More accurate responses

Testing RAG:

• Use "Compare AI vs RAG Chatbot" feature
• Test queries about your documents
• See difference in responses
• Verify accuracy

Best Practices

✅ Keep Documents Updated

• Refresh documents when content changes
• Rebuild vector store after major updates

✅ Monitor Performance

• Test chatbot responses
• Verify accuracy
• Rebuild if performance degrades

✅ Quality Matters

• Better documents = better vector store
• Well-structured content works best
• Remove outdated or irrelevant content

Account Considerations

All Account Types

• Can build vector stores
• Same functionality
• Processing time similar

Professional Accounts

• May have priority processing
• Can handle larger vector stores
• More storage capacity

Technical Details

How It Works

Embedding Model:

• Uses advanced AI models
• Converts text to numerical vectors
• Captures semantic meaning
• Understands relationships

Storage:

• Stored efficiently
• Fast retrieval
• Scalable to large datasets
• Optimized for search

Retrieval:

• Semantic similarity search
• Finds relevant chunks
• Ranks by relevance
• Returns top matches

Next Steps

After vector store is built:

1. Compare AI vs RAG - See the difference 2. Deploy Your Chatbot - Go live with RAG 3. Test Queries - Try questions about your documents 4. Monitor Performance - Check response accuracy

Your vector store is the foundation of RAG - build it carefully and maintain it well!