moshushi/big-rag • LM Studio HubUsage Examples
This document provides practical examples of using the Big RAG Plugin with different types of document collections.
Example 1: Technical Documentation Library
Scenario
You have a large collection of technical documentation, API references, and tutorials that you want to query using natural language.
Setup
Configuration
Usage Examples
This document provides practical examples of using the Big RAG Plugin with different types of document collections.
Example 1: Technical Documentation Library
Scenario
You have a large collection of technical documentation, API references, and tutorials that you want to query using natural language.
Setup
Configuration
~/Documents/tech-libraryVector Store Directory: ~/.lmstudio/tech-library-dbChunk Size: 1024 (larger chunks for technical content)Chunk Overlap: 200Retrieval Limit: 7Affinity Threshold: 0.6Max Concurrent Files: 5OCR: DisabledQuery 1: "How do I connect to PostgreSQL using Python?"
- Retrieves relevant passages from both Python and PostgreSQL documentation
- Combines information from multiple sources
- Provides code examples if available in the docs
Query 2: "What are the differences between var, let, and const in JavaScript?"
- Finds relevant sections from JavaScript documentation
- Returns explanations with examples
- May include best practices if documented
Query 3: "Show me MongoDB aggregation pipeline examples"
- Retrieves MongoDB-specific documentation
- Includes practical examples
- May reference multiple documents showing different use cases
Example 2: Research Paper Collection
You're a researcher with hundreds of PDF papers that you want to search and reference.
# Directory structure
~/Research/papers/
├── machine-learning/
│ ├── deep-learning/
│ ├── reinforcement-learning/
│ └── nlp/
├── computer-vision/
└── robotics/
- Documents Directory:
~/Research/papers
- Vector Store Directory:
~/.lmstudio/research-db
- Chunk Size: 768 (balanced for academic writing)
- Chunk Overlap: 150
- Retrieval Limit: 10 (more results for research)
- Affinity Threshold: 0.55
- Max Concurrent Files: 3 (PDFs are slower to parse)
- OCR: Disabled (assuming text-based PDFs)
Query 1: "What are the latest approaches to attention mechanisms in transformers?"
- Searches across all NLP papers
- Retrieves relevant sections discussing attention
- May include citations from multiple papers
Query 2: "Compare different reinforcement learning algorithms for robotics"
- Finds content from both RL and robotics papers
- Retrieves comparative information
- Provides context from multiple sources
Query 3: "What datasets are commonly used for object detection?"
- Searches computer vision papers
- Lists datasets mentioned in papers
- May include performance benchmarks
Example 3: Legal Document Archive
A law firm with thousands of legal documents, contracts, and case files.
# Directory structure
~/Legal/documents/
├── contracts/
│ ├── 2020/
│ ├── 2021/
│ ├── 2022/
│ ├── 2023/
│ └── 2024/
├── case-files/
└── regulations/
- Documents Directory:
~/Legal/documents
- Vector Store Directory:
~/.lmstudio/legal-db
- Chunk Size: 512 (standard for legal text)
- Chunk Overlap: 100
- Retrieval Limit: 5
- Affinity Threshold: 0.7 (higher precision for legal)
- Max Concurrent Files: 2 (careful processing)
- OCR: Enabled (for scanned documents)
Query 1: "Find all contracts with non-compete clauses"
- Searches across all contract documents
- Retrieves sections containing non-compete language
- Provides file references for review
Query 2: "What are the standard terms for intellectual property rights?"
- Finds IP-related clauses across documents
- Shows variations in language
- Helps identify standard vs. custom terms
Query 3: "Show precedents for breach of contract cases"
- Searches case files
- Retrieves relevant case information
- Provides context for similar situations
Example 4: Personal Knowledge Base
Personal collection of notes, articles, ebooks, and saved web pages.
# Directory structure
~/Knowledge/
├── books/
│ ├── fiction/
│ └── non-fiction/
├── articles/
│ ├── saved-webpages/
│ └── pdfs/
├── notes/
│ ├── work/
│ └── personal/
└── recipes/
- Documents Directory:
~/Knowledge
- Vector Store Directory:
~/.lmstudio/knowledge-db
- Chunk Size: 512
- Chunk Overlap: 100
- Retrieval Limit: 5
- Affinity Threshold: 0.5
- Max Concurrent Files: 4
- OCR: Enabled (for recipe images, etc.)
Query 1: "What did I save about productivity techniques?"
- Searches across articles and notes
- Finds productivity-related content
- Combines information from multiple sources
Query 2: "Find that recipe for chocolate cake"
- Searches recipe directory
- May use OCR if recipe is an image
- Returns recipe details
Query 3: "What books have I read about history?"
- Searches book collection
- Identifies history-related books
- May provide summaries or key points
Example 5: Software Development Project
Large codebase with documentation, README files, and code comments.
# Directory structure
~/Projects/myapp/
├── docs/
│ ├── api/
│ ├── guides/
│ └── tutorials/
├── README.md
├── CONTRIBUTING.md
└── src/
└── (various .md files for documentation)
- Documents Directory:
~/Projects/myapp
- Vector Store Directory:
~/.lmstudio/myapp-docs-db
- Chunk Size: 768
- Chunk Overlap: 150
- Retrieval Limit: 6
- Affinity Threshold: 0.55
- Max Concurrent Files: 5
- OCR: Disabled
Query 1: "How do I set up the development environment?"
- Finds setup instructions from README or guides
- Provides step-by-step information
- May reference multiple documentation files
Query 2: "What's the API for user authentication?"
- Searches API documentation
- Retrieves authentication-related endpoints
- Shows usage examples
Query 3: "How do I contribute to this project?"
- Finds CONTRIBUTING.md content
- Provides guidelines and workflow
- May include code style requirements
Example 6: Medical/Healthcare Records (Anonymized)
Healthcare provider with anonymized patient records, research notes, and medical literature.
# Directory structure
~/Medical/data/
├── research/
├── literature/
└── case-studies/
- Documents Directory:
~/Medical/data
- Vector Store Directory:
~/.lmstudio/medical-db
- Chunk Size: 512
- Chunk Overlap: 100
- Retrieval Limit: 8
- Affinity Threshold: 0.65 (higher precision for medical)
- Max Concurrent Files: 3
- OCR: Enabled (for scanned records)
Query 1: "What are common treatments for condition X?"
- Searches medical literature and case studies
- Retrieves treatment protocols
- Provides evidence-based information
Query 2: "Find cases with similar symptoms"
- Searches case studies
- Identifies similar presentations
- Helps with differential diagnosis
Large Text Collections (>10GB)
- Use higher concurrency (5-8)
- Disable OCR unless needed
- Consider processing in batches
- Use SSD for vector store
- Lower concurrency (2-3)
- Increase chunk size (1024+)
- Allow more time for initial indexing
- Monitor memory usage
- Enable OCR selectively
- Use moderate concurrency (3-4)
- Adjust threshold based on quality
- Test with small subset first
Frequently Updated Collections
- Enable auto-reindex
- Use file watching (future feature)
- Keep vector store on fast storage
- Regular maintenance
Example: No Results for Known Content
Problem: Querying for content you know exists returns no results.
- Lower affinity threshold (try 0.3-0.4)
- Rephrase query to match document language
- Check that file was actually indexed
- Verify file format is supported
Example: Too Many Irrelevant Results
Problem: Getting too many low-quality matches.
- Increase affinity threshold (try 0.7-0.8)
- Reduce retrieval limit
- Use more specific queries
- Adjust chunk size for content type
Problem: Initial indexing taking too long.
- Reduce max concurrent files
- Disable OCR if not needed
- Process subdirectories separately
- Check disk I/O performance
- Start Small: Test with a subset before indexing everything
- Tune Settings: Adjust based on your specific content
- Monitor Performance: Watch memory and disk usage
- Regular Maintenance: Periodically rebuild index for optimization
- Backup: Keep backups of your vector store
- Document: Note what settings work best for your use case
- Iterate: Refine queries and settings based on results
# Directory structure
~/Documents/tech-library/
├── python/
│ ├── official-docs/
│ ├── tutorials/
│ └── api-reference/
├── javascript/
│ ├── mdn-docs/
│ └── frameworks/
└── databases/
├── postgresql/
└── mongodb/
~/Documents/tech-libraryVector Store Directory: ~/.lmstudio/tech-library-dbChunk Size: 1024 (larger chunks for technical content)Chunk Overlap: 200Retrieval Limit: 7Affinity Threshold: 0.6Max Concurrent Files: 5OCR: DisabledQuery 1: "How do I connect to PostgreSQL using Python?"
- Retrieves relevant passages from both Python and PostgreSQL documentation
- Combines information from multiple sources
- Provides code examples if available in the docs
Query 2: "What are the differences between var, let, and const in JavaScript?"
- Finds relevant sections from JavaScript documentation
- Returns explanations with examples
- May include best practices if documented
Query 3: "Show me MongoDB aggregation pipeline examples"
- Retrieves MongoDB-specific documentation
- Includes practical examples
- May reference multiple documents showing different use cases
Example 2: Research Paper Collection
You're a researcher with hundreds of PDF papers that you want to search and reference.
# Directory structure
~/Research/papers/
├── machine-learning/
│ ├── deep-learning/
│ ├── reinforcement-learning/
│ └── nlp/
├── computer-vision/
└── robotics/
- Documents Directory:
~/Research/papers
- Vector Store Directory:
~/.lmstudio/research-db
- Chunk Size: 768 (balanced for academic writing)
- Chunk Overlap: 150
- Retrieval Limit: 10 (more results for research)
- Affinity Threshold: 0.55
- Max Concurrent Files: 3 (PDFs are slower to parse)
- OCR: Disabled (assuming text-based PDFs)
Query 1: "What are the latest approaches to attention mechanisms in transformers?"
- Searches across all NLP papers
- Retrieves relevant sections discussing attention
- May include citations from multiple papers
Query 2: "Compare different reinforcement learning algorithms for robotics"
- Finds content from both RL and robotics papers
- Retrieves comparative information
- Provides context from multiple sources
Query 3: "What datasets are commonly used for object detection?"
- Searches computer vision papers
- Lists datasets mentioned in papers
- May include performance benchmarks
Example 3: Legal Document Archive
A law firm with thousands of legal documents, contracts, and case files.
# Directory structure
~/Legal/documents/
├── contracts/
│ ├── 2020/
│ ├── 2021/
│ ├── 2022/
│ ├── 2023/
│ └── 2024/
├── case-files/
└── regulations/
- Documents Directory:
~/Legal/documents
- Vector Store Directory:
~/.lmstudio/legal-db
- Chunk Size: 512 (standard for legal text)
- Chunk Overlap: 100
- Retrieval Limit: 5
- Affinity Threshold: 0.7 (higher precision for legal)
- Max Concurrent Files: 2 (careful processing)
- OCR: Enabled (for scanned documents)
Query 1: "Find all contracts with non-compete clauses"
- Searches across all contract documents
- Retrieves sections containing non-compete language
- Provides file references for review
Query 2: "What are the standard terms for intellectual property rights?"
- Finds IP-related clauses across documents
- Shows variations in language
- Helps identify standard vs. custom terms
Query 3: "Show precedents for breach of contract cases"
- Searches case files
- Retrieves relevant case information
- Provides context for similar situations
Example 4: Personal Knowledge Base
Personal collection of notes, articles, ebooks, and saved web pages.
# Directory structure
~/Knowledge/
├── books/
│ ├── fiction/
│ └── non-fiction/
├── articles/
│ ├── saved-webpages/
│ └── pdfs/
├── notes/
│ ├── work/
│ └── personal/
└── recipes/
- Documents Directory:
~/Knowledge
- Vector Store Directory:
~/.lmstudio/knowledge-db
- Chunk Size: 512
- Chunk Overlap: 100
- Retrieval Limit: 5
- Affinity Threshold: 0.5
- Max Concurrent Files: 4
- OCR: Enabled (for recipe images, etc.)
Query 1: "What did I save about productivity techniques?"
- Searches across articles and notes
- Finds productivity-related content
- Combines information from multiple sources
Query 2: "Find that recipe for chocolate cake"
- Searches recipe directory
- May use OCR if recipe is an image
- Returns recipe details
Query 3: "What books have I read about history?"
- Searches book collection
- Identifies history-related books
- May provide summaries or key points
Example 5: Software Development Project
Large codebase with documentation, README files, and code comments.
# Directory structure
~/Projects/myapp/
├── docs/
│ ├── api/
│ ├── guides/
│ └── tutorials/
├── README.md
├── CONTRIBUTING.md
└── src/
└── (various .md files for documentation)
- Documents Directory:
~/Projects/myapp
- Vector Store Directory:
~/.lmstudio/myapp-docs-db
- Chunk Size: 768
- Chunk Overlap: 150
- Retrieval Limit: 6
- Affinity Threshold: 0.55
- Max Concurrent Files: 5
- OCR: Disabled
Query 1: "How do I set up the development environment?"
- Finds setup instructions from README or guides
- Provides step-by-step information
- May reference multiple documentation files
Query 2: "What's the API for user authentication?"
- Searches API documentation
- Retrieves authentication-related endpoints
- Shows usage examples
Query 3: "How do I contribute to this project?"
- Finds CONTRIBUTING.md content
- Provides guidelines and workflow
- May include code style requirements
Example 6: Medical/Healthcare Records (Anonymized)
Healthcare provider with anonymized patient records, research notes, and medical literature.
# Directory structure
~/Medical/data/
├── research/
├── literature/
└── case-studies/
- Documents Directory:
~/Medical/data
- Vector Store Directory:
~/.lmstudio/medical-db
- Chunk Size: 512
- Chunk Overlap: 100
- Retrieval Limit: 8
- Affinity Threshold: 0.65 (higher precision for medical)
- Max Concurrent Files: 3
- OCR: Enabled (for scanned records)
Query 1: "What are common treatments for condition X?"
- Searches medical literature and case studies
- Retrieves treatment protocols
- Provides evidence-based information
Query 2: "Find cases with similar symptoms"
- Searches case studies
- Identifies similar presentations
- Helps with differential diagnosis
Large Text Collections (>10GB)
- Use higher concurrency (5-8)
- Disable OCR unless needed
- Consider processing in batches
- Use SSD for vector store
- Lower concurrency (2-3)
- Increase chunk size (1024+)
- Allow more time for initial indexing
- Monitor memory usage
- Enable OCR selectively
- Use moderate concurrency (3-4)
- Adjust threshold based on quality
- Test with small subset first
Frequently Updated Collections
- Enable auto-reindex
- Use file watching (future feature)
- Keep vector store on fast storage
- Regular maintenance
Example: No Results for Known Content
Problem: Querying for content you know exists returns no results.
- Lower affinity threshold (try 0.3-0.4)
- Rephrase query to match document language
- Check that file was actually indexed
- Verify file format is supported
Example: Too Many Irrelevant Results
Problem: Getting too many low-quality matches.
- Increase affinity threshold (try 0.7-0.8)
- Reduce retrieval limit
- Use more specific queries
- Adjust chunk size for content type
Problem: Initial indexing taking too long.
- Reduce max concurrent files
- Disable OCR if not needed
- Process subdirectories separately
- Check disk I/O performance
- Start Small: Test with a subset before indexing everything
- Tune Settings: Adjust based on your specific content
- Monitor Performance: Watch memory and disk usage
- Regular Maintenance: Periodically rebuild index for optimization
- Backup: Keep backups of your vector store
- Document: Note what settings work best for your use case
- Iterate: Refine queries and settings based on results
# Directory structure
~/Documents/tech-library/
├── python/
│ ├── official-docs/
│ ├── tutorials/
│ └── api-reference/
├── javascript/
│ ├── mdn-docs/
│ └── frameworks/
└── databases/
├── postgresql/
└── mongodb/