User Guide

How to use the NorthBuilt RAG System effectively.

Table of Contents

  1. Getting Started
    1. Accessing the System
    2. First Query
  2. How to Ask Questions
    1. Best Practices
    2. Example Queries
    3. Client Filtering
  3. Understanding Responses
    1. Response Structure
    2. Source Citations
    3. When No Results Are Found
  4. Conversation Features
    1. Follow-Up Questions
    2. Starting Fresh
  5. Data Sources
    1. Fathom (Meeting Transcripts)
    2. HelpScout (Support Conversations)
    3. Linear (Projects)
  6. Tips for Teams
    1. Engineering Teams
    2. Sales Teams
    3. Support Teams
    4. Marketing Teams
  7. Troubleshooting
    1. Common Issues
    2. Getting Help
  8. Privacy and Security
    1. What Data Is Accessible
    2. Authentication
    3. Data Handling
  9. Keyboard Shortcuts
  10. Frequently Asked Questions

Getting Started

Accessing the System

  1. Navigate to the web application URL provided by your administrator
  2. Click “Sign in with Google”
  3. Complete the Google authentication flow
  4. You will be redirected to the chat interface

First Query

Once signed in, you can immediately start asking questions:

  1. Type your question in the input field
  2. Press Enter or click Send
  3. Wait 2-5 seconds for the response
  4. Review the answer and source citations

How to Ask Questions

Best Practices

Be Specific

Instead of: “What happened in meetings?”

Try: “What did we discuss with Acme Corporation about their pricing concerns?”

Include Context

  • Mention client names when relevant
  • Reference time periods if applicable
  • Include project names for context (displayed in sources)

Ask One Thing at a Time

Complex multi-part questions may yield less focused answers. Break them into separate queries.

Example Queries

Good Query Why It Works
“What were the action items from the last Acme meeting?” Specific client and clear request
“What support issues has Globex reported about the API?” Client + topic + context
“Summarize our discussions with Acme about enterprise pricing” Client + topic + action
Query to Improve Better Version
“meetings” “What was discussed in recent client meetings?”
“what about acme” “What are the key points from our Acme discussions?”
“help” “What support tickets mention authentication issues?”

Client Filtering

The system automatically detects client names in your questions and filters results accordingly. This means:

  • “What did Acme say about pricing?” retrieves only Acme-related documents
  • All projects under Acme are included for complete context
  • You don’t need to select a client from a dropdown

If the system cannot determine which client you mean (e.g., multiple clients with similar names), it will ask for clarification.


Understanding Responses

Response Structure

Each response includes:

  1. Answer: The generated response based on your documents
  2. Sources: The documents used to generate the answer
  3. Relevance Scores: How well each source matched your query

Source Citations

Sources are displayed with:

Field Description
Document Number Reference number ([1], [2], etc.)
Snippet Relevant excerpt from the document
Relevance Score 0.0-1.0 indicating match quality
Metadata Source type, client, project, date

Example:

[1] (0.92) Meeting with Acme - we discussed the enterprise pricing tier...
    Source: fathom | Client: Acme Corporation | Project: Enterprise Deal

When No Results Are Found

If the system cannot find relevant documents, it will:

  1. Indicate that no matching documents were found
  2. Suggest rephrasing your question
  3. Ask if you meant a different client

This is better than providing an incorrect or hallucinated answer.


Conversation Features

Follow-Up Questions

You can ask follow-up questions that reference previous context:

First query: “What did we discuss with Acme about pricing?”

Follow-up: “What were their concerns about the enterprise tier?”

The system maintains context within a session, so “their” refers to Acme.

Starting Fresh

To start a new, unrelated conversation:

  • Refresh the page, OR
  • Clear your query and mention a different client

Sessions maintain context for approximately 24 hours.


Data Sources

The system includes documents from multiple sources:

Fathom (Meeting Transcripts)

  • Video meeting transcripts
  • Automatically captured and indexed
  • Includes speaker identification
  • Searchable by client and meeting content

HelpScout (Support Conversations)

  • Customer support tickets
  • Email conversations
  • Includes full thread history
  • Searchable by client and issue content

Linear (Projects)

  • Used for entity registry (client/project names)
  • Teams in Linear map to clients
  • Projects in Linear provide context

Tips for Teams

Engineering Teams

  • Search for technical discussions: “What architecture decisions were made for the API?”
  • Find implementation details: “How did we implement authentication for Acme?”
  • Reference past decisions: “What was the rationale for using S3 instead of DynamoDB?”

Sales Teams

  • Research client history: “What has Acme expressed interest in?”
  • Prepare for meetings: “Summarize our relationship with Globex”
  • Find pricing discussions: “What pricing options did we discuss with Acme?”

Support Teams

  • Find similar issues: “Have other clients reported API timeout issues?”
  • Research client context: “What customizations does Acme have?”
  • Check past resolutions: “How did we resolve the authentication issue for Globex?”

Marketing Teams

  • Gather testimonials: “What positive feedback have clients given?”
  • Research use cases: “How is Acme using our product?”
  • Identify trends: “What features do clients request most?”

Troubleshooting

Common Issues

“No relevant documents found”

  • Try rephrasing your question
  • Check if you spelled the client name correctly
  • Use alternative client names or aliases

“Which client did you mean?”

  • The system found multiple matching clients
  • Select the correct one from the options provided
  • Be more specific in future queries

Slow response times

  • Normal response time is 2-5 seconds
  • First query after inactivity may be slower (cold start)
  • Very complex queries may take longer

Session expired (401 error)

  • Refresh the page to get a new authentication token
  • Sign in again if prompted

Rate limit exceeded (429 error)

  • Wait a moment and try again
  • Reduce query frequency if making many requests

Getting Help

If you encounter issues not covered here:

  1. Check with your administrator
  2. Review system status for any outages
  3. Provide the error message when reporting issues

Privacy and Security

What Data Is Accessible

  • Only documents associated with known clients
  • Documents are filtered by client based on your query
  • All projects under a client are accessible for complete context

Authentication

  • Access requires Google authentication
  • Sessions expire after inactivity
  • Each user has their own session

Data Handling

  • Queries are processed in AWS (us-east-1)
  • Responses are not stored beyond the session
  • Source documents remain in their original location

Keyboard Shortcuts

Shortcut Action
Enter Submit query
Shift+Enter New line in query
Ctrl/Cmd+K Focus query input
Escape Clear current input

Frequently Asked Questions

Q: How current is the data?

A: Documents are updated within 5-10 minutes of being added or modified in source systems (Fathom, HelpScout).

Q: Can I search across all clients?

A: If you don’t mention a specific client, the system may search broadly or ask for clarification. For best results, specify the client.

Q: Why don’t I see certain documents?

A: Documents must be properly tagged with client metadata. If a document appears to be missing, check with your administrator.

Q: Can I upload my own documents?

A: Currently, documents are ingested from integrated sources (Fathom, HelpScout). Direct upload is not supported via the web UI.

Q: How accurate are the responses?

A: Responses are generated based on retrieved documents. Relevance scores indicate confidence. Always verify critical information against source documents.

Q: Can I export my conversation?

A: Conversation export is not currently supported. You can copy/paste responses manually.


Last updated: 2026-01-01