Evaluate Hybrid Search for Legal Question-Answering using Qdrant & BM25/mxbai
VerifiedEvaluates hybrid search on legal QA data using Qdrant with BM25 and embeddings.
What this workflow does
This automation runs hybrid retrieval against a pre-indexed Qdrant collection and measures top-1 accuracy for legal question-answering tasks.
It targets developers and researchers who need to benchmark retrieval quality before building RAG applications.
Who is this for?
Legal AI developers and RAG engineers using n8n who need to test retrieval quality on domain-specific datasets before production deployment.
What problem it solves
Manually evaluating hybrid search (BM25 + semantic) performance on legal QA data is complex and time-consuming, making it hard to measure hits@1 before building a chatbot.
Live workflow preview
Interactive canvas of every node and connection — scroll and click to explore. Powered by n8n's preview.
Open the template on n8n to import and run it. View source template →
What it automates
Pre-RAG validation
Run the workflow on a LegalQAEval subset to confirm whether hybrid retrieval with RRF returns correct answer chunks in the top result.
Provider comparison
Switch between Qdrant Cloud Inference and an external embedding API to measure any change in hits@1 on the same collection.
Baseline benchmarking
Establish a reproducible hits@1 score for a simple BM25 + mxbai hybrid setup before adding rerankers or filters.
How the workflow works
The 1 nodes in this automation, in order.
- 1HTTP RequesthttpRequest
Apps & integrations used
How to set up Evaluate Hybrid Search for Legal Question-Answering using Qdrant & BM25/mxbai
- 1Import the workflow JSON into n8n and connect your Qdrant Cloud credentials via HTTP Request nodes.
- 2Ensure the collection created by the Part 1 Indexing workflow already exists.
- 3Choose embedding method: keep Qdrant Inference or replace the embedding call with your provider.
- 4Trigger the workflow on the evaluation question set from LegalQAEval.
- 5Review the hits@1 calculation output node for retrieval quality metrics.
- 6Store or export the evaluation results for later comparison.
How to customize this workflow
- →Replace Qdrant Inference with OpenAI embeddings by editing the HTTP Request node as shown in Part 1.
- →Increase the evaluation subset size by adjusting the dataset loader parameters.
- →Add a reranking step after RRF using an additional HTTP Request node.
- →Change the trigger from manual to scheduled to run periodic quality checks.
Evaluate Hybrid Search for Legal Question-Answering using Qdrant & BM25/mxbai: pros & cons
Pros
- +Ready-made evaluation of hits@1 on legal data
- +Demonstrates RRF fusion without custom code
- +Works with existing Qdrant collection from Part 1
- +Clear path to reuse Query Points node for RAG
Cons
- –Requires completion of the separate Indexing workflow first
- –Only evaluates a small subset by default
- –No built-in reranking or advanced fusion options
Frequently asked questions
It runs hybrid search (BM25 + mxbai) on a LegalQAEval subset stored in Qdrant and calculates hits@1 to measure retrieval quality.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…