Benchmark Report

WaveflowDB Benchmarking: Enterprise Retrieval Reinvented

Redefining accuracy, scalability, and cost-efficiency with full-corpus reranking.

Executive Summary

WaveflowDB delivers an enterprise-grade retrieval engine that eliminates traditional top-k constraints by applying full-corpus reranking across all documents.

MS MARCO Benchmark Results

  • MRR @ 161,536 documents: 0.55
  • Top-2 average result position
  • End-to-end API suite for retrieval, reranking, embeddings, and monitoring
  • Predictable, linear cost scaling versus the multi-vendor, multi-pipeline RAG stack

This positions WaveflowDB as a single-source, unified retrieval layer built for modern, agentic enterprise workflows.

Why Enterprises Choose WaveflowDB

Full-Corpus Reranking
Ensures relevant results surface—even deep in large corpora.
Unified Retrieval Stack
Storage, embeddings, vector search, reranking, and monitoring in one API suite.
Linear, Predictable Cost Model
No hidden inference or reranking overhead.
Enterprise-Grade Scalability
Maintains precision even at 150k+ document scale.

Waveflow — Suite of Platform Capabilities

Waveflow-UI

No-code agent builder UI

Waveflow-DB

Serverless vector database

Waveflow-Rerank

Enterprise-class reranker

Waveflow-Framework

Multi-agent workflows

Waveflow-Observability

Full-stack monitoring

MS MARCO Benchmark Overview
Dataset
MS MARCO Passage Ranking (Dev)
Corpus Size
161,536 documents
Queries Evaluated
19,697
MRR Result
0.55

Result: MRR = 0.55 → Relevant answer appears in top 2 positions on average

Why Full-Corpus Reranking Matters

Traditional retrieval stacks rely on top-k filtering, assuming relevant answers lie in the first 100 documents.

This is rarely true in enterprise data.

WaveflowDB performs global scoring across the full corpus, eliminating retrieval bias and significantly enhancing precision.

Impact: No more "missed documents" due to aggressive filtering.

MRR Comparison

MethodMRR@100Notes
BM250.18Lexical limits
DPR Dense0.38Bi-encoder
ColBERT0.349Late interaction
MonoT5/DuoT50.39Cross-encoder
WaveflowDB0.55 @161,536 docsFull-corpus

Architecture & API Suite

WaveflowDB APIs
waveflow-doc-search
waveflow-upload-batch
waveflow-update-batch
waveflow-reranking
waveflow-meta
waveflow-conversational-AI

End-to-end simplicity—no multi-service orchestration required.

TCO & Cost Advantages

Traditional RAG Stack

Pinecone + Cohere + LLM + OpenSearch

  • ✗High infra costs, reranking fees
  • ✗Complex orchestration
  • ✗Unpredictable scaling
WaveflowDB
  • Flat, predictable pricing
  • No per-query rerank fees
  • Single unified stack
  • Lower operational load

Business Leader Viewpoint

For CEOs
  • Reduced TCO
  • Faster strategic insights
  • Higher accuracy → better customer experiences
For CTOs
  • Scalable, storage-native architecture
  • Model-agnostic embeddings
  • Unified monitoring and metering
Key Takeaways
  • Retrieval precision: Top-2 accuracy across full corpus
  • Unified RAG stack: Storage, embeddings, reranking, search, observability
  • Scalable architecture with linear pricing
  • Purpose-built for enterprise and agentic workflows