Skip to main content
CAIL Sandbox Docs

Grounding AI with Knowledge Collections

After uploading files (PDFs, Markdown, plain text) to a knowledge collection, custom models can retrieve relevant passages from those documents in response to situated tasks or course-specific questions.

When a student asks, “What does the syllabus say about late submissions?”, an uploaded syllabus gives the model a document to search and cite. Without that document, the model may generate a confident but incorrect response. Test whether it retrieves the relevant passage and represents your policy accurately before students rely on it.

Creating a Knowledge Base

  1. Request Workspace access by emailing the CUNY AI Lab team. Once enabled, click Workspace in the left sidebar.
  2. Select Knowledge
  3. Click Create beside the Workspace tabs
    • A dialog opens asking you to describe what you are building
  4. Give it a name
    • Use a name your students or colleagues will recognize, such as “ENG 2100 Fall 2026 Readings” or “IRB Protocol Archive”
  5. Describe its purpose
    • Name the materials the collection contains and the tasks it should support, such as checking course policies or comparing research methods.
  6. Set access
    • Keep Private while building. Use Add Access to grant your course group Read access.
    • Choose Public only for materials intended for all signed-in Sandbox users, where permitted.
  7. Click Create Knowledge

Uploading Documents

  1. Drag and drop files into the knowledge base, or click to browse
    • Supported formats. PDF, Markdown, plain text
    • You can upload multiple files at once
  2. Wait for processing to complete
    • The system splits your documents into chunks and creates searchable embeddings. Processing time depends on file size and the configured services.

After processing finishes, attach the collection to the custom model that should use it.

Connecting to a Model

  1. Go to Workspace > Models and edit the model you want to ground
  2. Scroll to the Knowledge section
  3. Select the knowledge base you just created
    • You can attach multiple knowledge bases to a single model so it can retrieve from those collections.
  4. Click Save & Update

The knowledge base is now attached to the model. Grant its intended users access to the knowledge base as well as the model so they can use it for retrieval.

Tip. Start with a small collection (syllabus + 2-3 key readings) to test how well the model retrieves and uses your materials. Add more documents once you are confident in the results.

Advanced Settings

View details

What Happens Under the Hood

When you upload a document, the Sandbox splits it into chunks and converts each chunk into a numerical representation called an embedding. These representations help it find related passages even when a question uses different words. A question about “thesis committee requirements” may retrieve a passage about “dissertation advisory boards”; check that the passage addresses the question you asked.

Retrieval depends on the model’s configuration. In native tool-calling mode, the model can use knowledge tools to fetch passages; other modes add retrieved context automatically. Check that it retrieves the relevant material when testing.

Choosing What to Upload

Use clean, well-structured text so the system can divide it into usable passages.

Works well

  • Markdown files and plain text
  • Well-formatted PDFs with clear headings and paragraphs
  • Course syllabi, handbooks, policy documents
  • Research papers and annotated bibliographies

May need preprocessing

  • Complex PDFs with multi-column layouts, tables, or embedded images
  • Scanned documents without OCR
  • Slide decks (convert to text or PDF with notes first)

If a PDF produces poor results, convert it to Markdown and check that headings and paragraphs remain in the intended order before uploading it again.

Managing Your Files

Access all uploaded files through Settings > Data Controls > Manage Files. This centralized manager lets you search by filename, sort by name or date, and inspect file metadata. Deleting a file here removes it from all knowledge bases and deletes the corresponding embeddings.

RAG Template (Admin)

Administrators can customize the template for retrieval-augmented generation (RAG) via Settings > Admin > Documents > RAG Template. Use it to tell the model how to cite retrieved passages and respond when those passages do not address the question.

Example for CUNY

You are assisting a CUNY researcher. Respond based primarily on
the provided context. When using information from documents,
indicate the source. If the context does not adequately address
the query, say so and suggest how the user might find additional
information. Prioritize accuracy over elaboration.

Embedding Model Configuration

Administrators can check or change the embedding model in Settings > Admin > Documents. After changing it, use Reindex to rebuild knowledge-base embeddings. Files uploaded directly to chats must be uploaded again; changing the setting does not rebuild them automatically.

Callout

For researchers. Build a knowledge collection from the studies or methods you want to compare, then attach it to a custom model. Ask the model to compare specific claims or methods and check its citations against the uploaded documents.

Additional Resources

← Return to Custom Models | Continue to Tools & Skills →