Security & Performance

Security & Performance

This guide covers important privacy considerations and performance optimization techniques for using Ollama in healthcare consulting work.

Privacy & Confidentiality

Ollama provides several key security features that make it ideal for healthcare consulting:

• In-house deployment: All interactions remain within Candesic’s secure infrastructure
• Client data protection: Safe to use with confidential client information and sensitive materials
• No external training: Your queries and uploaded documents are not used to train future models
• Project isolation: Create separate chat threads for different client projects to maintain confidentiality

Consulting Advantage

Unlike public models like ChatGPT, Ollama allows you to work with confidential healthcare data, financial models, and proprietary client information without data privacy concerns.

Document Handling Best Practices

Supported Document Types

• PDFs: Research reports, market analyses, regulatory documents
• Text files: Clean data extracts, notes, interview transcripts
• CSVs: Structured data, financial metrics, market statistics
• Images: Charts, graphs, presentation slides (with Gemma3 Vision)

Preparation Tips

• Excel data: Export to CSV with clear headers
• Large documents: Break into logical sections
• Visual data: Provide context descriptions
• OCR limitations: Verify text extraction from images

Project-Specific Handling

• Confidential materials: Create dedicated knowledge collections for sensitive client information
• Access control: Use custom models to control which team members can access specific knowledge collections
• Data lifecycle: Remove sensitive data from collections once projects are completed
• Classification: Tag documents appropriately with security level indicators

Optimizing Performance

Model Selection for Workload

• Use Llama 3.2 for routine tasks: Faster and optimized for multiple users during high-demand project periods
• Schedule intensive tasks: For complex analyses, consider using models during off-peak hours
• Match model to task: Use specialized models only when their capabilities are needed
• Batch similar requests: Group related queries together for more efficient processing

Content Preparation Techniques

• Pre-prepare templates: Create reusable prompt templates for common consulting deliverables
• Batch document uploads: Process multiple reports together for integrated analysis
• Save successful prompts: Document effective prompts in your team’s knowledge base for reuse
• Optimize for context window: Break large documents into logical sections with appropriate context

For Time-Critical Projects

During tight client deadlines, be aware that model performance may vary with usage volume. Schedule complex analyses during off-peak hours when possible and allocate sufficient time for iterative refinement of outputs.

Implementation Best Practices

1. Create project spaces: Organize your work by creating separate chat threads for each client project
2. Use consistent naming: Name chats and knowledge collections clearly for easy identification
3. Document methodology: Record your approach to using AI for specific analyses for quality assurance
4. Respect model limitations: Verify critical facts and figures, especially financial calculations

User Permissions

Candesic maintains a role-based security model for the Ollama platform:

• Standard access: All consultants can use base models and public knowledge collections
• Team access: Project teams can access project-specific collections
• Admin access: Limited to authorized personnel for system management

Getting Support

For security questions or performance optimization assistance:

• Contact the Candesic technology team via the internal support channel
• Reference the internal knowledge base for additional security guidelines
• Report any unusual system behavior or potential security concerns immediately