Instructor guide
Build a course TA around your materials
Instructors control the course knowledge base, bot behavior, approved student access, and review tools for each selected course.
Recommended Setup Workflow
1. Select the course
Use the course selector before editing the bot. All dashboard actions apply to the currently selected course.
2. Build or copy the course setup
Upload new materials or use Copy course setup to reuse bot behavior and knowledge-base documents from another course you teach.
3. Build and preview modules
Add module materials, activities, and a few practice questions. Publish the intended modules and materials, then use Student preview to check the learning path. Activities have their own publication setting.
4. Edit bot behavior
Set expectations for tone, boundaries, academic integrity, and how the TA should handle uncertainty. The tutor uses the current module or lesson as context while retaining access to the full course knowledge base.
5. Check reliability
Use Reliability tools and Test bot with realistic student questions before releasing the TA to students.
6. Add students
Use Manage student list to approve student emails for the selected course. Create a registration code for each unregistered student and share it privately.
7. Review and maintain
Use analytics and students' chat history to improve documents, instructions, and starter questions over time. Personal activity completion is not a grade or a measure of mastery.
Dashboard Features
Course modules
Organize a learning sequence with module overviews, activities, and instructor-written ungraded practice questions. Attach multiple existing knowledge-base files to each module's Course materials section without creating activities or copying files.
Student preview
Open Student preview from Course modules to see published lessons, materials, and practice questions. You can also start Check your understanding, submit a test answer, revise once, and finish the check using the same materials and feedback rules as students. Test responses stay in temporary memory, expire after 30 minutes of inactivity or a server restart, and never enter student records or analytics. Use Exit student preview to return to editing.
Data analysis snapshot
Use the main dashboard for a quick course check: students included in analysis, active students, weekly usage, response quality, and high-level instructional signals.
Advanced data analysis
Open a deeper course-level analysis page for conversation depth, usage rhythm, knowledge-base match quality, topic-by-question-type patterns, feedback reasons, anonymized examples, engagement, conversation quality, knowledge transfer, and instructional improvement indicators.
Test bot
Try the selected course TA as an instructor. Test chats are not saved to student chat history, so this is the safest place to check new materials or prompt changes.
Students' chat history
Review student conversations for the selected course, including source-match indicators below assistant responses. This can help identify confusing assignments, recurring policy questions, or topics that need class clarification.
Edit bot behavior
Adjust the TA's tone, boundaries, academic integrity rules, and conversation starters. This shapes how the TA responds, while the knowledge base supplies course facts.
Knowledge base
Upload course materials, assign categories, rename documents, add optional notes telling the bot when and how to use each source, download original files, and remove outdated documents.
Copy course setup
Reuse bot behavior, conversation starters, and knowledge-base documents from another course you teach. You can append copied documents or replace the selected course knowledge base first.
Reliability tools
Test retrieval for sample questions, inspect source matches, review instructor source notes, preview excerpts, and check knowledge-base health issues before students rely on the TA.
Practice question review
Review anonymously labeled student-generated practice questions, refine strong examples, add notes, and hide weak examples so future practice-question generation better reflects your course expectations.
Manage student list
Approve student emails and create private, one-time registration codes for new students. View, filter, and sort analysis status. Ask a system admin to exclude test accounts from analytics and research exports; system admins can change Include in analysis for the selected course. Exclusion keeps login and chat history, applies to saved and future activity, and persists through roster changes. Open analysis tabs refresh after changes or when you return to them; Usage rhythm counts only included chats. This setting does not replace research consent. Broader user and instructor management stays in admin mode.
Admin users
Visible only to admin accounts. Admins can create courses and manage approved student, instructor, and admin accounts across courses.
Account settings
Update account information and sign out when needed.
Switch to another course
Move between courses when assigned to more than one course TA.
Using Data For Teaching And Research
The dashboard is not only a usage counter. It helps instructors understand how students seek support, where they struggle, and which course materials may need clarification. When handled responsibly, collected data can support better teaching decisions and meaningful learning research.
The main dashboard is designed as a quick Data Analysis Snapshot. It highlights course activity, weekly use, response quality, and a small set of high-level instructional signals so instructors can see what deserves attention without reading every detailed chart.
Use Advanced data analysis when you want a deeper view of conversation depth, usage rhythm, knowledge-base match quality, topic-by-question-type patterns, feedback reasons, engagement, conversation quality, knowledge transfer, and improvement signals. Feedback reasons are summarized by the most common reasons first, with an option to expand the full list when more reasons exist. These are course-level indicators generated from saved interactions. They are meant to guide instruction and research planning, not to grade students or evaluate individual performance.
Improve teaching
Use the snapshot for a quick read, then use advanced topic trends, question types, and topic-by-question-type patterns to identify concepts, assignments, or policies that need clearer instruction.
Monitor support quality
Review helpfulness ratings, unresolved feedback, and response coverage to spot answers that may need source updates or behavior-prompt refinement.
Strengthen the knowledge base
Use retrieval confidence and knowledge-base match quality to find materials that are missing, too broad, outdated, or categorized incorrectly.
Understand engagement
Use weekly usage, usage rhythm, active student counts, and conversation depth to understand when students seek help and whether they are asking quick or sustained questions.
Inform research
Use de-identified aggregate patterns to study help seeking, learning friction, engagement, and the instructional role of AI-supported learning assistants.
How the main indicators are calculated
Active students: Students who started at least one saved chat, compared with approved students.
Repeat use: Active students with more than one saved chat, compared with all active students.
Follow-up rate: Saved chats with two or more student messages, showing whether students continue the conversation after an initial question.
Conversation quality: Uses follow-up rate, average student-message length, extended conversations, feedback requests, and very short inputs as practical indicators of interaction depth.
Knowledge transfer: Looks for application-oriented language such as apply, workplace, organization, case, scenario, transfer, or implementation.
Improvement signals: Highlights low or missing retrieval confidence, selected issue feedback, and topics that may need clearer knowledge-base materials.
Research use should rely on de-identified or aggregated records whenever possible. Student names, emails, and direct identifiers should be excluded from research reporting unless a separate approved process permits otherwise. Dashboard patterns should be interpreted with course timing, assignment due dates, Canvas activity, and instructor observations; they should not be used as grades or formal student performance judgments.
Start With These Materials
A simple setup is enough to make the TA useful quickly. Focus on materials students would normally check in Canvas before asking the instructor.
- Syllabus
- Assignment instructions
- Sample assignments or model excerpts, if appropriate
- Rubrics
- Weekly lecture slides
- Key handouts or study guides
- Course policy documents
How Materials Shape The Bot
The TA retrieves relevant excerpts from uploaded documents and combines them with the behavior prompt. If the knowledge base is incomplete, outdated, or unclear, the bot may provide less useful guidance or rely on general explanation.
- Upload materials before students begin using the TA.
- Keep documents aligned with the current semester.
- Remove outdated materials.
- Include rubrics and assignment instructions when students will ask assignment questions.
- Add short bot-use notes for sources that have special boundaries, such as drafts, examples, policies, or materials that students should confirm in Canvas.
- Test the bot after major uploads or behavior-prompt changes.
Preparing Files For Better Answers
The TA works best when uploaded files are current, readable, and specific. Think of the knowledge base as the course context the TA is allowed to rely on.
The Knowledge Base page supports PDF, DOCX, PPTX, Excel, text, Markdown, CSV, and JSON uploads. You can upload multiple files at once. If you provide one shared title for multiple files, the app saves them as Title (1), Title (2), and so on. After upload, use Rename when a document title needs to be clearer.
In Saved documents, search by document name, original source filename, or notes to the bot. Filter by category and sort by name, date added, or most recently updated. Clear filters to see all documents again. Filtering only changes your view; the TA still has access to the entire course knowledge base.
If a review suggestion appears, choose Show documents to review to find the affected files. Each document explains the reason and what to do next. A General category can be kept when appropriate; otherwise choose a more specific category, which saves automatically. For limited extracted text, compare the preview with your original file. Short excerpts are valid. If content is missing, upload a searchable version (use OCR for scanned files) or paste the relevant text. Verify the new preview before removing the old document.
Use descriptive file names that include the week, topic, or assignment name.
Prefer PDF, DOCX, PPTX, or text files with selectable text rather than scanned images.
Use document categories intentionally: syllabus, assignments/projects, slides, readings/concepts, course policy, or general.
Rename uploaded documents when a file name is not clear enough for instructors or source chips.
Add optional notes to tell the bot when a source should be used, when it should be treated cautiously, or when students should verify details elsewhere.
Separate outdated materials from current-semester materials.
Upload assignment instructions and rubrics together when possible.
Avoid uploading private student work unless it is approved, anonymized, and appropriate for course support.
Remove or replace files when deadlines, requirements, or policies change.
Knowledge Base Notes And Categories
Categories help the TA choose the right source before it answers. Syllabus and policy questions prefer syllabus or policy documents, assignment questions prefer assignments/projects, slide questions prefer slides, and concept questions prefer readings/concepts. Use General only when a document does not fit a narrower category.
Optional notes to the bot are useful when a source needs extra guidance. For example, you can tell the TA to use a rubric for grading criteria, use a sample assignment only as an example, or remind students to confirm changing dates and announcements in Canvas. These notes are included in retrieval and reliability previews, but they are not displayed to students as standalone instructions.
Bot Behavior
Use Edit bot behavior to define tone, academic integrity expectations, course boundaries, and how the TA should use uploaded materials. Keep the prompt concise. The knowledge base should carry course facts; the behavior prompt should guide how the TA responds.
Useful behavior-prompt details
Tell the TA how concise it should be, when to refer students back to Canvas or the instructor, how to handle assignment questions, and what it should refuse to do. Avoid putting long syllabus content in the behavior prompt; upload that content to the knowledge base instead.
Testing Before Students Use It
Use Test bot after uploading files or changing the behavior prompt. Try questions that students are likely to ask about deadlines, assignment expectations, rubric criteria, and difficult concepts.
Use Reliability tools for a deeper check. The reliability page shows detected category, category search order, source-match confidence, retrieved source names, instructor notes, and excerpt previews. If source confidence is low or none, add clearer documents, improve categories, rename unclear sources, or add a short note telling the bot where that source should be used.
If the TA gives a vague answer, upload a clearer document or revise the behavior prompt. If it gives an answer that is too long, tighten the behavior prompt. If it misses course-specific details, the knowledge base probably needs more direct source material.
Practice Question Quality Review
Students can intentionally generate practice questions from their Learning Snapshot. The app now grounds those questions in matching course knowledge-base materials when available, then saves the generated set for instructor quality review under an anonymous student label.
Use Practice question review to inspect the original questions, source-match confidence, and retrieved source labels. You can revise strong examples, add an internal note, or hide weak examples from future guidance. Revised examples are reused as instructor-refined examples when later students generate practice questions for the same topic.
This creates a lightweight quality loop: students receive study prompts, instructors keep the prompts aligned with the course, and the system gradually learns from instructor-approved examples without changing the course knowledge base or exposing student names.
Student Attachments
Students can attach PDF, DOCX, PPTX, and Excel files during chat for one-time help with a specific document. These attachments are not added to the course knowledge base. They are temporary session materials, retained only long enough to support follow-up questions in that chat and then removed based on the configured temporary attachment retention window.
Official course materials should still be uploaded by the instructor through the Knowledge Base page. This keeps course-level knowledge under instructor control.
Semester Maintenance
Weekly
Upload new slides, handouts, or clarifications. Categorize new documents and test one or two likely student questions in Reliability tools.
Before major assignments
Upload the assignment prompt, rubric, examples, and any instructor guidance students should use. Add bot-use notes if the source has special limitations.
After recurring confusion
If many students ask the same question, update the knowledge base or clarify the assignment instructions.
End of term
Review whether old materials should be archived, removed, or updated before the next course offering.
Canvas Integration Status
The current app does not connect to Canvas, request Canvas API tokens, or automatically import Canvas records. Course modules and materials in Virtual TA are managed separately.
Upload course materials to the knowledge base and attach them to modules. Publishing controls student browsing and downloads; the AI tutor can use all knowledge-base materials in the selected course, including unpublished files. Unpublishing is not a way to keep information out of AI answers.
Tell students where to submit graded work and check official grades, such as Canvas. Virtual TA provides a course learning workspace with materials, practice, and tutoring; it does not synchronize those items with Canvas or provide a gradebook or graded-submission workflow. Keep separately maintained course information consistent.
Instructor Responsibility
The Virtual TA supports instruction, but it does not replace instructor judgment. Continue to direct students to Canvas, the syllabus, official announcements, and course staff for final guidance on deadlines, grades, policies, and assignment expectations.