Knowledge Base

Overview

The Knowledge Base is your organization’s central document repository. ADM uses it for retrieval-augmented generation (RAG) to produce accurate, contextual answers grounded in your policies, procedures, and technical docs.

Purpose

• Centralize organization-specific knowledge.
• Enrich AI responses with verified context.
• Share domain knowledge and maintain a single source of truth.

How it works

1. Upload documents.
2. Index content (text extraction and structuring).
3. Embed content into vectors for fast similarity search.
4. Retrieve relevant sections at question time.
5. Answer with sources and citations.

When you ask a question, ADM automatically searches the Knowledge Base, pulls the most relevant passages, and cites them in the response.

Uploading Documents

Navigate

• Main menu: Knowledge Base.
• Command: type /knowledge-base in any conversation.

Upload

1. Click Upload Document (or drag and drop files).

2. (Optional) Add metadata:

   • Title, Description, Tags, Access Level.

3. Click Upload and wait for indexing to complete.

Supported formats

• Fully supported: PDF, DOC/DOCX, TXT, MD, XLS/XLSX (Spreadsheets index static values; formulas are not executed.)
• Partially supported: PPTX (text only), JSON (experimental), XML (experimental)
• Not supported: Images, audio, video, archives, executables

Limits

• Max file size: 25 MB
• PDF: ≤ 500 pages
• Word: ≤ 200 pages

Document Indexing

What indexing does

• Extracts text and structure.
• Splits content into meaningful chunks.
• Generates semantic embeddings.
• Stores vectors for fast similarity search.

Status indicators

• Indexing ⟳: Processing
• Indexed ✓: Ready to use
• Failed ✗: Needs attention

Re-indexing

• Occurs automatically when you upload a new version.
• Trigger manually with Re-index if results look off or after system updates.

Using Knowledge in Conversations

Automatic context

• ADM analyzes your question, retrieves relevant passages (typically 3–5), synthesizes an answer, and shows citations.

Citations include

• Document name, and page numbers when available.
• Click a citation to open the source and verify the context.

See sources

• Expand Sources at the end of a response to view all referenced documents.

Managing Your Knowledge Base

View

• The document table shows Name, Upload Date, Size, Status, Tags. Sort and filter as needed.

Search

• Search by Name, Tags, Content, and Upload Date. Combine criteria to narrow results; save frequent searches.

Edit metadata

1. Open a document > Edit.
2. Update Title, Description, Tags, Access.
3. Save Changes. (No re-index required for metadata edits.)

Delete

1. Select document(s) > Delete > confirm Delete Permanently.

   Warning This action cannot be undone. Download a copy first if needed.

Organize with tags

• Use clear, consistent tags (e.g., policy, technical-doc, onboarding, engineering, project-alpha).
• Establish a simple tag taxonomy for teams to follow.

Best practices

• Keep documents current; review quarterly.
• Remove superseded content.
• Use descriptive filenames (e.g., Dept_DocType_Version_Date).
• Monitor which docs are cited often to spot gaps.

Technical Architecture (Brief)

• RAG pipeline: vector store + semantic retrieval + grounded generation.
• Vector database: FAISS for fast similarity search.
• Embeddings: transformer models create semantic vectors.
• Chunking: sentence/token strategies preserve context within optimal chunk sizes.
• Retrieval: top matches (cosine similarity) provided as context to the model for faithful answers.

Quality controls include checks for relevancy, contextual precision, faithfulness, and hallucination/toxicity safeguards.

Integration with External Knowledge Sources

ADM supports external sources via Model Context Protocol (MCP):

• Confluence: query live pages without manual upload.
• Jira: reference tickets and project artifacts.
• Future options: Google Drive, SharePoint, and more.

Note MCP integrations require administrator configuration.

Troubleshooting

Document not appearing

• Verify format and size limits; refresh the page.
• Wait for Indexing to finish.
• Known: PPTX may not list correctly in some cases; contact support if persistent.

Incorrect/incomplete answers

• Confirm the source document contains the correct content.
• Re-index the document.
• Check for conflicting versions or duplicates.

Indexing failed

• Open the file locally to confirm it isn’t corrupted.
• Re-upload in a supported format and within limits.

Slow indexing

• Large/complex docs take longer.
• Break very large documents into smaller files if needed.

Security and Privacy

Access Control

The Knowledge Base respects your organization's security requirements:

• Document-Level Permissions: Set access levels during upload to control who can view and query each document. Options typically include Public (all users), Department-specific, and Private (uploader only).
• Role-Based Access: Your ADM administrator can configure role-based access control (RBAC) to automatically apply permissions based on user roles.
• Audit Logging: All document uploads, accesses, and deletions are logged for security and compliance purposes.

Data Protection

Your documents are protected through multiple security layers:

• Encryption at Rest: All documents are encrypted in storage using industry-standard encryption protocols.
• Encryption in Transit: Data transfers between your browser and ADM servers use TLS encryption.
• Data Isolation: In multi-tenant environments, your documents are isolated from other organizations' data.
• Retention Policies: Configure automatic deletion of documents after a specified period to comply with data retention requirements.

Compliance

The Knowledge Base supports various compliance frameworks:

• GDPR compliance for European data
• SOC 2 Type II certification
• ISO 27001 information security standards
• HIPAA compliance for healthcare documents (when configured)