AI Quality Engineering / RAG Evaluation
Retrieval Quality Platform

RAG Evaluation Workbench

Measure retrieval quality before generation reaches production. Benchmark lexical, semantic, and hybrid retrieval with labelled relevance sets and reproducible IR evidence.

KeylessBrowser-nativeReproducible

Evaluate the retrieval layer independently

Run BM25, MiniLM semantic embeddings, or hybrid retrieval against a controlled corpus. Inspect ranked evidence, compare standard IR metrics, and export a traceable evidence artifact with SHA-256 provenance.

BM25Transformers.jsMiniLMRecall@KMRRnDCG@KSHA-256 evidence
Execution100% browser-side
API keysNone required
Retrievers3 modes
EvidenceExportable JSON

Benchmark workspace

Configure corpus, labels, retrieval mode and scoring parameters

Evaluation data

Edit the corpus and relevance labels, or run the included reference benchmark.

Retriever configuration

Semantic mode downloads a compact Hugging Face ONNX model into the browser cache on first use.

Ready. BM25 runs entirely offline after page load.

Evaluation summary

Aggregate metrics from the latest benchmark run
Recall@K
Precision@K
MRR
nDCG@K
Hit Rate
P50 latency

Per-query metrics

Inspect query-level retrieval quality and ranking behavior.

Run a benchmark to populate results.

Ranked retrieval evidence

Review the retrieved chunks, scores, source document and relevance label behind each metric.

Top-ranked chunks and relevance labels appear here.

Metric guide

How to interpret the benchmark metrics used by this workbench.

MetricQuestion it answers
Precision@KOf the top-K retrieved documents, how many are relevant?
Recall@KOf all labelled relevant documents, how many appeared in the top K?
MRRHow early does the first relevant document appear?
nDCG@KHow good is the ranking, with stronger reward for relevant results near the top?
Hit RateDid retrieval find at least one relevant document in the top K?
P50 / P95 latencyWhat retrieval latency did the benchmark observe?

Architecture

The benchmark evaluates document-level retrieval quality even when retrieval operates on chunks.

Corpus + labelled relevance set │ ▼ Chunking configuration │ ┌─────┼───────────┐ ▼ ▼ ▼ BM25 MiniLM Hybrid │ │ │ └─────┼───────────┘ ▼ Ranked document evidence │ ▼ Precision@K · Recall@K · MRR · nDCG · Hit Rate · latency │ ▼ Evidence JSON + SHA-256 provenance