pdf-rag

name	pdf_rag
description	PDF RAG (Retrieval-Augmented Generation) skill for searching reference documents. Use when needing specifications, tactical procedures, equipment details, or any information from uploaded PDF manuals and guides.

PDF RAG Search Skill

Overview

This skill enables semantic search across uploaded PDF documents including manuals, field guides, intelligence reports, and tactical references. Use it to augment video analysis with contextual information.

When to Use

• Looking up equipment specifications (range, speed, capabilities)
• Finding tactical procedures and doctrine
• Cross-referencing video observations with documentation
• Answering "why" or "how" questions beyond visual analysis
• Getting identification guides for detected objects

Usage

result = pdf_rag_skill(query="tank operational range capabilities")

Parameters

• query (required): Semantic search query describing needed information
• pdf_sources (optional): List of specific PDF filenames to search
• top_k (optional): Number of results to return (default: 5)

Result Structure

Returns a dictionary containing:

• status: "success", "no_results", or "error"
• result: Contains search results with:
  • results: List of matching passages with:
    • rank: Position in relevance ranking
    • text: Extracted text chunk
    • relevance_score: Similarity score (0.0-1.0)
    • source: PDF filename
  • sources_used: List of PDFs that had matches
  • num_results: Number of results found

Relevance Score Guidelines

• > 0.7: Highly relevant, strong match
• 0.5 - 0.7: Moderately relevant
• < 0.5: Weakly relevant, may be off-topic

Best Practices

Good Query Construction

# GOOD - specific, technical queries
pdf_rag_skill("T-72 tank identification features")
pdf_rag_skill("convoy spacing tactical formation")
pdf_rag_skill("tank thermal signature detection range")

# BAD - too vague
pdf_rag_skill("information about tanks")  # Too general
pdf_rag_skill("how many tanks in video")  # Use videodb_query instead

Combining with Video Analysis

# Step 1: Detect objects in video
tanks = videodb_query_skill("object_search", object_type="tank")

# Step 2: Get reference information
specs = pdf_rag_skill("tank capabilities and weaknesses")

# Step 3: Synthesize in your response

Error Handling

Always check status:

result = pdf_rag_skill("tank specifications")

if result["status"] == "success":
    # Use results
    for item in result["result"]["results"]:
        print(f"[{item['source']}] {item['text'][:200]}...")
elif result["status"] == "no_results":
    # Normal - not all queries match content
    print("No relevant information found in PDFs")
elif result["status"] == "error":
    # System error
    print(f"Error: {result['message']}")

Source Attribution

Always cite sources when using PDF information:

result = pdf_rag_skill("convoy tactics")
if result["status"] == "success":
    for r in result["result"]["results"]:
        # Include source in your response
        print(f"According to {r['source']}: {r['text']}")