Welcome to Dexrag

Dexrag is a Monte Carlo RAG (Retrieval-Augmented Generation) service that solves the fundamental limitations of embedding-based document retrieval.

Why Dexrag?

Google DeepMind research has demonstrated that embeddings hit mathematical limits at scale, achieving less than 20% recall on the LIMIT benchmark. Dexrag uses probabilistic exploration to break these ceilings.

Key Features

• Monte Carlo Search: Explores documents intelligently like AlphaGo explores chess moves
• Adaptive Learning: Personalizes to your users' patterns and learns domain-specific terminology
• No Scaling Limits: Unlike embeddings which cap at 250M documents for 4096 dimensions
• Cost Effective: $99/mo vs $500+/mo for vector databases
• Explainable Results: Shows exploration tree instead of black box similarity scores

Performance

• 15% better than GPT-4 embeddings on query #1
• 47% better by day 30 with adaptive learning
• 89% recall on DeepMind LIMIT benchmark vs <20% for embeddings

Getting Started

Ready to break free from embedding limitations? Check out our Getting Started guide to begin.