Checking…

Eval-first RAG infrastructure

Build RAG systems you can actually measure.

Retrieve better context. Generate grounded answers. Know when your system should refuse — with evaluation, traces, and CI gates built in.

Query pipeline

  1. Query

  2. Rewrite

  3. Hybrid retrieval

  4. RRF

  5. Rerank

  6. Ground

  7. Answer

Why ContextIQ

Typical RAG
ContextIQ
Build → hope
Eval → build → measure
Vector search only
Dense + BM25 + RRF + rerank
Always answers
Grounded refusal
Manual testing
Golden evaluation set
No regression protection
CI quality gates
Black box
Full pipeline traces

Evaluation snapshot

75 golden queries · structural · hybrid_rerank

Full dashboard →

Context Recall

89.3%

How often the relevant source information was successfully retrieved.

Context Precision

41.6%

Share of retrieved chunks that are actually relevant to the question.

Faithfulness

100%

Evaluated with extractive generation

Answer Relevancy

68.7%

How well the answer addresses the user's question.

Refusal Accuracy

100%

Correctly refusing when the corpus cannot support an answer.

End-to-end pipeline

  1. 01

    Ingest

    Fetch public docs into a versioned raw zone.

  2. 02

    Parse

    Preserve headings, tables, lists, and code structure.

  3. 03

    Chunk

    Structural chunking tuned for technical documentation.

  4. 04

    Embed

    Dense vectors (BGE-small) with optional hash fallback.

  5. 05

    Understand query

    Route, rewrite, and optional HyDE expansion.

  6. 06

    Hybrid retrieve

    Dense cosine + BM25 sparse search in parallel.

  7. 07

    Generate

    Extractive, Ollama, or Bedrock grounded generation.

  8. 08

    Ground / refuse

    Cite sources or refuse when evidence is thin.

Stack

PythonRAG pipeline & API
Next.jsProduct UI
TypeScriptFrontend
BGE-smallEmbeddings
BM25Sparse retrieval
PostgreSQLOptional pgvector store
RailwayAPI hosting
VercelUI hosting
GitHub ActionsCI & eval gate

See how an answer was produced.

Open RAG playground