Examples

LlamaFarm Examples Skill

Reference for LlamaFarm example projects and their patterns.

Command Entry Point

/llamafarm:example                       # List all available examples
/llamafarm:example quick_rag             # Scaffold from quick_rag example
/llamafarm:example fda_rag --name myproj # Scaffold with custom name
/llamafarm:example gov_rag --show        # Show example config without scaffolding

When to Load

Load this skill when:

• User runs /llamafarm:example command
• User asks about example projects
• User wants to scaffold from examples
• User needs working configuration references
• Learning LlamaFarm patterns

Example Catalog

Implementation

Step 1: List Available Examples (no arguments)

Available LlamaFarm Examples
============================

quick_rag
  Minimal RAG setup for getting started quickly.
  Documents: 2 markdown files about AI/ML
  Features: Basic similarity search, Universal Runtime integration
  Best for: Learning LlamaFarm basics

fda_rag
  FDA regulatory document analysis.
  Documents: FDA correspondence letters (PDFs)
  Features: Entity extraction, semantic chunking, citation tracking
  Best for: Regulatory/legal document analysis

gov_rag
  Government and municipal document processing.
  Documents: City ordinances, planning documents
  Features: Large document handling, hierarchy extraction
  Best for: Large complex PDFs, municipal docs

ocr_and_document
  Document OCR and image text extraction.
  Documents: Scanned PDFs, images with text
  Features: OCR pipeline, form parsing, document extraction
  Best for: Scanned documents, image-based PDFs

rag_pipeline
  Advanced RAG configuration showcase.
  Documents: Mixed format technical docs
  Features: Multiple databases, reranking, hybrid search
  Best for: Learning advanced RAG patterns

Usage:
  /llamafarm:example <name>           # Scaffold from example
  /llamafarm:example <name> --show    # Preview config only

Step 2: Show Example (--show flag)

When user runs /llamafarm:example quick_rag --show:

# Read the example's llamafarm.yaml
cat ~/workspace/pivot/llamafarm/examples/quick_rag/llamafarm.yaml

Present the configuration with annotations explaining key choices.

Step 3: Scaffold Project

When user runs /llamafarm:example quick_rag:

# Import example using CLI
lf examples import quick_rag --name quick_rag

# Or manually copy files
mkdir -p ./quick_rag
cp -r ~/workspace/pivot/llamafarm/examples/quick_rag/* ./quick_rag/
cd ./quick_rag

Report results:

Project Scaffolded: quick_rag
=============================

Created in: ./quick_rag/

Files:
  llamafarm.yaml    - Project configuration
  files/            - Sample documents
  README.md         - Example documentation

Next Steps:

1. Navigate to project:
   cd quick_rag

2. Start services:
   lf start

3. Create and process dataset:
   lf datasets create -s markdown_processor -b main_db research
   lf datasets upload research ./files/*
   lf datasets process research

4. Chat with your documents:
   lf chat "What are neural scaling laws?"

Requirements:
  - Universal Runtime running on port 11540

Step 4: Custom Project Name

When user runs /llamafarm:example fda_rag --name my_legal_project:

1. Copy example files to ./my_legal_project/
2. Update name field in llamafarm.yaml
3. Report new project location

Progressive Disclosure

For specific example patterns, load:

• quick-rag.md - Minimal RAG setup
• fda-rag.md - FDA regulatory analysis
• gov-rag.md - Government document processing

Scaffolding Commands

# List examples
lf examples list

# Import example
lf examples import quick_rag --name my-project

# Show example config
cat examples/quick_rag/llamafarm.yaml

Example Locations

Examples are located in the LlamaFarm repository:

llamafarm/examples/
├── quick_rag/
│   ├── llamafarm.yaml
│   ├── manifest.yaml
│   ├── README.md
│   └── files/
├── fda_rag/
├── gov_rag/
├── ocr_and_document/
└── rag_pipeline/

Common Patterns Across Examples

All examples include:

• Complete llamafarm.yaml configuration
• Sample documents in files/ directory
• manifest.yaml for import metadata
• README.md with usage instructions
• run_example.sh for interactive walkthrough

Standard workflow:

1. Initialize/import example
2. Start services
3. Create dataset
4. Upload and process files
5. Query with chat

Learning Path

Beginner

1. Start with quick_rag

   • Minimal configuration
   • 2 markdown files
   • Basic similarity search
   • Universal Runtime integration

2. Learn the workflow:

   lf start
   lf datasets create -s markdown_processor -b main_db research
   lf datasets upload research ./files/*
   lf datasets process research
   lf chat "What is the document about?"
   

Intermediate

3. Try fda_rag for PDFs

   • PDF parsing with fallbacks
   • Entity extraction
   • Semantic chunking
   • Larger documents

4. Explore gov_rag for scale

   • Large document handling
   • Hierarchy extraction
   • Geospatial entities

Advanced

5. Study rag_pipeline for optimization

   • Multiple databases
   • Hybrid search
   • Reranking
   • Strategy composition

6. Explore ocr_and_document for vision

   • OCR pipelines
   • Image processing
   • Document extraction

Example Details

quick_rag

Purpose: Minimal viable RAG for learning

Configuration highlights:

• Single Universal Runtime model (unsloth/Qwen3-4B-GGUF:Q4_K_M)
• ChromaStore vector database
• Basic similarity retrieval
• Markdown parsing with heading extraction

Sample files:

• neural_scaling.md - AI research notes
• engineering_practices.md - Software engineering tips

Requirements:

• Universal Runtime installed and running
• Model auto-downloaded on first use

fda_rag

Purpose: Regulatory document analysis

Configuration highlights:

• PDF parsing with entity extraction
• Semantic chunking (1200 chars / 150 overlap)
• Organizations, dates, products extraction
• Citation-aware system prompt

Sample files:

• Multiple FDA warning letters (PDF)
• Regulatory correspondence

Requirements:

• Universal Runtime or OpenAI for chat
• PDF dependencies (PyPDF2, LlamaIndex)

gov_rag

Purpose: Municipal and government documents

Configuration highlights:

• Large document handling
• Hierarchy extraction (outline/headings)
• Geospatial entity extraction (GPE, FAC, LOC)
• Table extraction

Sample files:

• City ordinances
• Planning documents
• Municipal codes

Requirements:

• Sufficient disk space for large vectors
• Consider Universal Runtime for faster embeddings

ocr_and_document

Purpose: Scanned document processing

Configuration highlights:

• OCR pipeline integration
• Image text extraction
• Form field parsing
• Multi-model with vision capability

Features:

• Surya OCR
• PaddleOCR
• Document extraction models

Requirements:

• Universal Runtime for OCR models
• Additional ML dependencies

rag_pipeline

Purpose: Advanced RAG showcase

Configuration highlights:

• Multiple vector databases
• Hybrid search (semantic + keyword)
• Cross-encoder reranking
• Metadata filtering

Demonstrates:

• Database routing
• Strategy composition
• Performance optimization

Requirements:

• Multiple model endpoints
• Larger resource allocation

Key Configuration Patterns

Minimal Valid Config (from quick_rag)

version: v1
name: my-project
namespace: default

runtime:
  models:
    - name: default
      provider: universal
      model: unsloth/Qwen3-4B-GGUF:Q4_K_M
      default: true

prompts:
  - name: default
    messages:
      - role: system
        content: You are a helpful assistant.

rag:
  default_database: main_db
  databases:
    - name: main_db
      type: ChromaStore
      embedding_strategies:
        - name: default_embeddings
          type: UniversalEmbedder
      retrieval_strategies:
        - name: basic_search
          type: BasicSimilarityStrategy
          default: true

  data_processing_strategies:
    - name: default
      parsers:
        - type: MarkdownParser_LlamaIndex
          file_include_patterns: ["*.md"]

datasets:
  - name: docs
    database: main_db
    data_processing_strategy: default

PDF Analysis (from fda_rag)

Key additions:

• PDFParser_LlamaIndex with semantic chunking
• EntityExtractor for ORG, DATE, PERSON
• Larger chunk sizes (1200/150)
• Citation-aware prompts

Large Documents (from gov_rag)

Key additions:

• HeadingExtractor with hierarchy
• TableExtractor for structured data
• Multiple parser fallbacks
• Increased chunk overlap

Advanced RAG (from rag_pipeline)

Key additions:

• Multiple databases with routing
• HybridUniversalStrategy
• CrossEncoderRerankedStrategy
• Metadata filtering