Introduction

Vecta is a RAG evaluation platform that helps you ship reliable AI systems. Test retrieval quality, catch hallucinations, and track performance over time.

Why Vecta?

Before Vecta:

• No way to know if your RAG system actually works
• Hallucinations discovered by users in production
• Can't tell if code changes improve or break retrieval

With Vecta:

• Automated benchmarks tailored to your data
• Precision/Recall/F1 scores for retrieval quality
• Accuracy and groundedness metrics for generation
• Track performance over time

Key Features

Granular Metrics

• Chunk-level: Did you retrieve the right paragraphs?
• Page-level: Did you find the right pages?
• Document-level: Did you find the right documents?

Three Evaluation Types

• Retrieval Only: Test search quality
• Generation Only: Test LLM responses
• Retrieval + Generation: Test your complete pipeline

Universal Compatibility

• Works with any vector database
• Flexible schema configuration
• Pre-built connectors for popular databases

Next Steps

• Quickstart → - Connect and evaluate in 5 minutes
• Data Sources → - Connect your database
• Benchmarks → - Create test datasets
• Evaluations → - Run and interpret tests