data-type-handling

Data Type Handling

Overview

Use this parent skill when source material is not plain prose or when different data types need different parsing, metadata, chunking, and retrieval strategies.

Problem Statement

Code, APIs, tables, images, diagrams, and mixed media lose important meaning when treated as generic text. RAG systems need data-specific handling to preserve symbols, structure, and visual context.

Key Concepts

• Code structure: Preserve symbols, paths, APIs, and examples.
• Multimodal extraction: Convert images, diagrams, and tables into retrievable representations.
• Metadata alignment: Keep source references consistent across modalities.
• Specialized retrieval: Match query type to the right content representation.

Implementation Guide

Step 1: Classify Source Content

Identify whether the corpus is code-heavy, image-heavy, table-heavy, or mixed.

Step 2: Preserve Native Structure

Use parsers and metadata that retain symbols, captions, headers, page numbers, and source locations.

Step 3: Choose Retrieval Representations

Index text, summaries, captions, code symbols, table schemas, or multimodal embeddings as appropriate.

When to Use This Skill

• Building RAG for technical documentation or code repositories
• Indexing PDFs, diagrams, screenshots, or tables
• Combining text and non-text evidence in answers

When NOT to Use This Skill

• The corpus is simple prose
• The main issue is vector database operations
• The main issue is production latency

Related Skills

• RAG for Code Documentation
• RAG for Multimodal Content

Metrics & Success Criteria

• Accurate retrieval of symbols, captions, tables, and source references
• Lower hallucination on code or visual questions
• Clear traceability back to original artifacts