ChatFDA RAG System Development

A skill for building and maintaining a RAG (Retrieval-Augmented Generation) system prototype that queries FDA data. Focuses on clean code organization, minimal implementation, and proper handling of FDA label chunks.

What This Skill Does

This skill guides AI agents to work on a RAG system for FDA data queries while following specific project conventions around code style, design principles, and domain-specific requirements for handling FDA labels and text chunks.

Instructions

Code Design Principles

1. **Simplicity First**

- Keep code simple, small, and concise

- Don't include behavior beyond what is being discussed

- Always produce minimal code

- Demonstrate good code organization and design

2. **Formatting Standards**

- Use 4 spaces indentation for TypeScript files

- Keep configuration files in their generated format (preserve original indentation and semicolons)

- Do not use semicolons in TypeScript code (but preserve them in config files)

3. **Comment Guidelines**

- Keep existing comments as they are unless code changes require updates

- Do not add comments describing what changed or referring to deleted code

- Use comments judiciously

- Prefer refactoring to smaller functions over adding explanatory comments

Domain-Specific Knowledge

**FDA Labels and Chunks:**

- Do NOT deduplicate chunks by label (chunks from the same label are likely different and both relevant)

- DO deduplicate or rank lower chunks with similar text from different manufacturers

- Challenge: Increase diversity in chunks while avoiding manufacturer duplicates

Implementation Approach

When working on this codebase:

1. Read existing code structure before making changes

2. Apply the minimal code principle - only add what's necessary

3. Maintain 4-space indentation in TypeScript files

4. Preserve formatting in configuration files

5. Factor complex logic into small, well-named functions rather than adding comments

6. When implementing chunk deduplication:

- Focus on cross-manufacturer similarity detection

- Preserve multiple chunks from the same label

- Prioritize chunk diversity in results

Example Usage

**User Request**: "Add a function to deduplicate FDA label chunks"

**Agent Response**:

Constraints