LAMB User Manual #

Audience: Educators and instructional designers who create and manage AI learning assistants using the LAMB platform. This manual covers everything from your first login to publishing assistants into your LMS.

1. Getting Started #

1.1 Logging In #

Open your LAMB instance URL in a browser. You will see the login screen.

Enter the email and password provided by your institution’s LAMB administrator, then click Login.

If your institution has enabled self-registration, click Sign up and complete the form. You will need the signup key provided by your organization administrator.

LTI Access: If your institution has configured LTI Creator integration, you can access LAMB directly from your LMS (Moodle, Canvas, etc.) without a separate login. Your account is created automatically on first launch.

1.2 The Dashboard #

After logging in, you land on the Dashboard — your home screen.

The dashboard shows:

• Your name, organization, and role (e.g., Admin, Member)
• My Resources — a summary of everything you own:
  • Assistants — how many you have created and how many are published
  • Knowledge Bases — your document collections
  • Rubrics — evaluation rubrics you have created
  • Prompt Templates — reusable prompt configurations
• Shared with Me — resources other educators in your organization have shared with you

1.3 Navigation #

The top navigation bar is always visible:

2. Learning Assistants #

Assistants are the core of LAMB — AI-powered tutors, evaluators, and learning companions that your students interact with.

2.1 Viewing Your Assistants #

Click Learning Assistants in the navigation bar. The My Assistants tab shows all assistants you own.

For each assistant, the list shows:

• Name and publication status (Published / Unpublished)
• Description
• Creation and last update dates
• Technical configuration — the prompt processor, connector, LLM model, and RAG processor
• Actions — View (eye icon), Export (download icon), Delete (trash icon)

You can search by name or description, filter by status (All, Published, Unpublished), and sort by different criteria.

2.2 Creating an Assistant #

Click the + Create Assistant button. The creation form appears.

Fill in the following fields:

Basic Information #

Configuration Panel (right side) #

Advanced Mode #

Toggle Advanced Mode to access additional settings like the Connector (OpenAI, Ollama, etc.) and Prompt Processor.

Import from JSON #

If you have an assistant configuration exported from another LAMB instance, click Import from JSON to load it.

Click Save when done. Your assistant is created in Unpublished state — only you can see and test it.

2.3 Writing Effective System Prompts #

The system prompt is the most important part of your assistant. Here are some tips:

• Define the role clearly: “You are a music history tutor for high school students.”
• Set the scope: “Focus on 1960s British rock, particularly the British blues revolution.”
• Give behavioral instructions: “Keep explanations clear, accurate, age-appropriate, and engaging.”
• Handle edge cases: “If the question is outside the provided context, say so plainly and answer cautiously.”
• Encourage interaction: “Ask brief follow-up questions when helpful.”

2.4 Viewing Assistant Details #

Click the View (eye) icon on any assistant to open its detail page.

The detail page has several tabs:

The Properties view shows all configuration details including the system prompt, prompt template, connected knowledge bases, and the full technical configuration (connector, LLM, RAG processor, etc.).

2.5 Editing an Assistant #

Click the Edit tab (or the Edit button on Properties) to modify your assistant.

The edit form is similar to the creation form. On the right panel you can:

• Change the LLM model, connector, and processors
• Enable or disable Vision Capability
• Select a different RAG Processor and configure RAG options
• Choose which Knowledge Bases to connect (checkboxes appear when a RAG processor is selected)

Click Save Changes when done. Click Cancel to discard changes.

2.6 Sharing Assistants #

The Share tab lets you grant other educators in your organization access to your assistant.

Click Manage Shared Users to add or remove users. Shared users can:

• View the assistant’s configuration
• Chat with the assistant
• See its test results

Shared users cannot edit, delete, or publish your assistant — only the owner can do that.

Note: Sharing must be enabled at the organization level by your administrator. If you don’t see the Share tab, contact your admin.

3. Knowledge Bases (RAG) #

Knowledge Bases let your assistants answer questions using your own documents — lecture notes, textbooks, articles, web content — instead of relying solely on the AI model’s general training data. This is called Retrieval-Augmented Generation (RAG).

3.1 Why Use a Knowledge Base? #

Without a Knowledge Base, your assistant answers from its general training data. This means:

• It might give outdated information
• It cannot reference your specific course materials
• It might “hallucinate” — generate plausible-sounding but incorrect answers

With a Knowledge Base, the assistant retrieves relevant passages from your documents and uses them as context when generating answers. This makes responses more accurate, grounded, and specific to your course.

3.2 Managing Knowledge Bases #

Navigate to Sources of Knowledge > Knowledge Bases.

The list shows all your knowledge bases with their name, creation date, sharing status, and actions (Share, View, Edit, Delete).

Click Create Knowledge Base to create a new one. Give it a descriptive name related to its content.

3.3 Adding Documents #

Click View on a knowledge base to see its details and manage files.

The detail page has three tabs:

Files Tab #

Shows all uploaded documents with their name, size, type, and processing status. Upload documents by dragging files or clicking the upload button. Supported formats include:

• PDF — textbooks, articles, handouts
• Markdown / Text — lecture notes, study guides
• Word documents — course materials

Each file goes through an ingestion pipeline that splits it into searchable chunks and creates vector embeddings for semantic search.

Ingest Content Tab #

Import content from external sources:

• Web pages — paste a URL to scrape and ingest its content
• YouTube videos — paste a YouTube URL to ingest the transcript

Query Tab #

Test your knowledge base by entering a search query. The system shows the most relevant chunks it would retrieve, along with similarity scores. This is useful for verifying that your documents are properly indexed and that the right content is being found.

3.4 Connecting a Knowledge Base to an Assistant #

To use a Knowledge Base with an assistant:

1. Edit your assistant
2. Set the RAG Processor to one of:
   • Simple Rag — basic retrieval, good for most use cases
   • Context Aware Rag — more sophisticated context handling
   • Rubric Rag — specialized for rubric-based evaluation
3. Set RAG Top K — how many document chunks to retrieve (1-10, default 3)
4. Check the Knowledge Bases you want to connect (they appear as checkboxes)
5. Ensure your Prompt Template includes the {context} placeholder (this is where retrieved content is inserted) and the {user_input} placeholder (where the student’s question goes)

Example Prompt Template:

Context:
{context}

Student question: {user_input}

Answer using the provided context. Cite your sources when possible.

3.5 Sharing Knowledge Bases #

Click the Share button to make a knowledge base available to other educators in your organization. Shared knowledge bases are read-only for other users — they can connect them to their assistants but cannot modify the content.

4. Rubrics (EvaluAItor) #

Rubrics let you create AI assistants that evaluate student work against structured criteria. Navigate to Sources of Knowledge > Rubrics.

4.1 What is EvaluAItor? #

EvaluAItor is LAMB’s rubric-based evaluation system. You define evaluation criteria with performance levels, and the AI assistant uses these criteria to assess student submissions. This is particularly useful for:

• Evaluating written assignments
• Assessing project presentations
• Providing structured feedback on student work
• Ensuring consistent evaluation across multiple submissions

4.2 Creating a Rubric #

Click Create Rubric and define:

• Rubric Name — a descriptive title
• Description — what this rubric evaluates
• Subject — the academic discipline
• Scoring Type — points or qualitative levels
• Maximum Score — the total possible score
• Criteria — each criterion has:
  • A name and weight (percentage of total)
  • Performance levels (e.g., Excellent, Good, Adequate, Insufficient) with descriptions and point values

4.3 Using a Rubric with an Assistant #

To create an evaluator assistant that uses a rubric:

1. Create a new assistant (or edit an existing one)
2. Set the RAG Processor to Rubric Rag
3. Select the rubric in the assistant configuration
4. Write a system prompt that instructs the assistant to evaluate submissions using the rubric
5. The prompt template should include {context} (where the rubric content is injected) and {user_input} (the student’s submission)

When a student submits their work, the assistant retrieves the rubric criteria and provides structured evaluation feedback based on the defined performance levels.

4.4 Example: Published Evaluator Assistant #

Here is an example of a published evaluator assistant with a rubric attached:

Notice the Selected Rubric section at the bottom showing the rubric content with its criteria and performance levels. The LTI Publish Details on the right show the integration information for connecting to your LMS.

5. Prompt Templates #

Prompt Templates let you save and reuse system prompt configurations across multiple assistants. Access them from the Prompt Templates tab on the Learning Assistants page.

5.1 Managing Templates #

• + New Template — create a template with a name, description, system prompt, and prompt template
• Edit — modify an existing template
• Share / Unshare — make a template available to other educators in your organization
• Duplicate — create a copy of a template
• Delete — remove a template

5.2 Using Templates #

When creating or editing an assistant, click Load Template next to the System Prompt field to populate the assistant’s configuration from a saved template. This is useful when you want to create multiple assistants with the same base personality or instructional approach.

6. Testing Your Assistant #

Before publishing an assistant to students, you should test it thoroughly. LAMB provides two ways to test: direct chat and structured test scenarios.

6.1 Direct Chat #

Click the Chat with [assistant name] tab on the assistant detail page to have a conversation with your assistant. This lets you quickly check:

• Does it respond appropriately to typical student questions?
• Does it stay within the defined scope?
• Does it use knowledge base content when available?

6.2 Debug Mode (Bypass) #

Debug mode shows you exactly what the AI model receives — the full system prompt, retrieved knowledge base content, and assembled prompt — without actually calling the AI model. This costs zero tokens and is invaluable for verifying that:

• The RAG pipeline is working (is {context} populated with relevant content?)
• The prompt template is correctly assembled
• The right knowledge base documents are being retrieved

6.3 Test Scenarios #

The Tests tab lets you create structured test scenarios and run them systematically.

Creating Scenarios #

Click + Add Scenario to create a test case:

• Title — a descriptive name (e.g., “Bluesbreakers overview”)
• Type — Normal, Edge case, or Adversarial
• Message — the test question or prompt
• Expected behavior — what a good response should include

Running Tests #

Test Results #

Test runs appear below the scenarios with:

• The model used and the date of the run
• Token count — how many tokens were consumed
• Response time — how long the model took to respond
• Evaluation — thumbs up (Good), thumbs down (Bad), or Mixed

Click on a run to see the full response. Click Evaluate to record your judgment.

Best practice: Always run Debug (bypass) first to verify your RAG pipeline is working correctly, then run with the real model.

7. Activity & Analytics #

The Activity tab shows usage statistics for your assistant.

You can see:

• Total chats — how many conversations have occurred
• Unique users — how many different students used the assistant
• Total messages — the total number of messages exchanged
• Messages per chat — the average conversation length

The Chat History table lists individual conversations with the date, user, title, and message count. Click on a chat to see the full conversation transcript (subject to privacy settings).

Use Filter Results to search by date range, user, or content.

8. Publishing to Your LMS #

Once you are satisfied with your assistant’s performance, you can publish it so students can access it.

8.1 Publishing an Assistant #

On the assistant’s Properties page, click the Publish button. This:

1. Registers the assistant as available for students
2. Generates LTI integration credentials (Consumer Key and Secret)
3. Creates a Tool URL for LMS integration

After publishing, the Properties page shows the LTI Publish Details panel:

• Assistant Name — the published name
• Model ID — the internal identifier
• Tool URL — the URL to configure in your LMS
• Consumer Key — the LTI OAuth key
• Secret — ask your LAMB administrator for the secret value

8.2 Connecting to Moodle (LTI) #

To make your assistant available in a Moodle course:

1. In LAMB: Publish the assistant and note the Tool URL and Consumer Key
2. In Moodle: Go to your course > Turn editing on > Add an activity > External Tool
3. Configure the External Tool:
   • Tool URL: paste the URL from LAMB
   • Consumer Key: paste from LAMB
   • Shared Secret: get this from your LAMB administrator
4. Save the activity

Students clicking the activity link in Moodle will be redirected to a chat interface where they can interact with your assistant.

Unified LTI (Recommended) #

If your institution has set up Unified LTI, a single LTI tool is configured for the entire LAMB instance. As an instructor, when you click the LTI link for the first time, you see a setup page where you choose which published assistants to make available in that activity. Students then see all selected assistants in one view.

The Unified LTI approach also provides an Instructor Dashboard with:

• Usage statistics (how many students, how many messages)
• Student access tracking
• Chat transcripts (if enabled during setup) showing student conversations with the assistants

8.3 Unpublishing #

Click the Unpublish button to remove the assistant from student access. Students will no longer be able to start new conversations, but existing chat history is preserved.

9. Collaboration #

9.1 Sharing Assistants #

Share your assistants with colleagues so they can learn from your configurations, test them, or use them as inspiration. See Section 2.6 for details.

9.2 Sharing Knowledge Bases #

Share your document collections so other educators can connect them to their own assistants. See Section 3.5 for details.

9.3 Shared Templates #

Share your prompt templates as reusable starting points for other educators. Access Shared Templates from the Prompt Templates tab.

9.4 The “Shared with Me” View #

The Shared with Me tab on the Learning Assistants page shows all assistants that other educators have shared with you. You can view their configuration and chat with them, but you cannot edit or publish them.

10. Tips and Best Practices #

For System Prompts #

• Be specific: “You are a biology tutor for 10th-grade students” is better than “You are a tutor”
• Set boundaries: Tell the assistant what it should and should not do
• Define tone: Specify if the assistant should be formal, friendly, Socratic, etc.
• Include language preferences: If your students work in a specific language, specify it

For Knowledge Bases #

• Quality over quantity: Well-structured documents produce better retrieval than large dumps of raw content
• Use clear headings and sections in your documents — they help the chunking process
• Test retrieval using the Query tab before connecting to an assistant
• Start with a few documents and add more as needed

For RAG Configuration #

• Always use a prompt template with {context} and {user_input} when RAG is enabled
• Test with bypass first to verify the pipeline before spending tokens on real AI calls
• Adjust RAG Top K — more chunks (5-10) give more context but may dilute relevance; fewer chunks (1-3) are more focused

For Testing #

• Create at least 3-5 test scenarios covering:
  • Normal questions — the core use case
  • Edge cases — related but off-center topics
  • Adversarial inputs — attempts to make the assistant go off-topic
• Run bypass first, then real tests
• Use the evaluation system to track quality over time

For Publishing #

• Test thoroughly before publishing
• Start with a small group of students if possible
• Monitor the Activity tab to spot issues early
• Iterate: use test results and student feedback to improve the system prompt and knowledge base

11. Glossary #

LAMB User Manual — Version 0.6 Last updated: April 2026