Enhanced

Legal Document RAG Pipeline

Purpose-built RAG pipeline for legal documents. High-precision retrieval with strong keyword matching for legal terminology, citations, and clause references.

text

Production

Run in Builder

from mixpeek import Mixpeek

client = Mixpeek(api_key="YOUR_API_KEY", namespace="legal-docs")

# 1. A bucket for contracts, and a collection that splits pages into layout blocks
bucket = client.buckets.create(
    bucket_name="contracts",
    bucket_schema={"properties": {"contract": {"type": "pdf"}}},
)
collection = client.collections.create(
    collection_name="contracts",
    source={"type": "bucket", "bucket_ids": [bucket["bucket_id"]]},
    feature_extractor={
        "feature_extractor_name": "document_graph_extractor",
        "version": "v1",
    },
)

# 2. Upload and process
client.buckets.upload(
    bucket["bucket_id"],
    blobs=[{"property": "contract", "type": "pdf", "data": "s3://your-bucket/contracts/msa-acme.pdf"}],
)
client.collections.trigger(collection["collection_id"])

# 3. Dense and BM25 search fused with RRF (BM25 reads the namespace text payload
#    indexes), then a cross-encoder for precision
retriever = client.retrievers.create(
    retriever_name="legal-search",
    collection_identifiers=["contracts"],
    input_schema={"query": {"type": "text", "required": True}},
    stages=[
        {
            "stage_name": "search",
            "stage_id": "feature_search",
            "parameters": {
                "searches": [
                    {
                        "feature_uri": "mixpeek://document_graph_extractor@v1/intfloat__multilingual_e5_large_instruct",
                        "query": {"input_mode": "text", "value": "{{INPUT.query}}"},
                        "top_k": 50,
                    },
                    {
                        "feature_uri": "mixpeek://document_graph_extractor@v1/intfloat__multilingual_e5_large_instruct",
                        "query": {"input_mode": "text", "value": "{{INPUT.query}}"},
                        "top_k": 50,
                        "lexical": True,
                    },
                ],
                "fusion": "rrf",
                "final_top_k": 50,
            },
        },
        {
            "stage_name": "rerank",
            "stage_id": "rerank",
            "parameters": {
                "inference_name": "BAAI__bge_reranker_v2_m3",
                "query": "{{INPUT.query}}",
                "document_field": "text_raw",
                "top_k": 10,
            },
        },
    ],
)

results = client.retrievers.execute(retriever["retriever_id"], inputs={"query": "Section 14.2 indemnification obligations"})
for doc in results["documents"]:
    print(doc["page_number"], doc["text_raw"][:120], doc["score"])

Feature Extractors

Document Graph Extractor

Decompose PDFs into spatial blocks (paragraphs, tables, forms, headers) with layout classification and E5 text embeddings.

Retriever Stages

feature search

Search and filter documents by vector similarity using feature embeddings

filter

rerank

Rerank documents using cross-encoder models for accurate relevance

sort

Frequently asked questions

Why is BM25 weight higher than vector weight?

Legal documents rely heavily on exact terminology and section references. Higher BM25 weight ensures precise term matching alongside semantic understanding.