🔁 ZeroEntropy Reranker for Elasticsearch

Boost your Elasticsearch results with LLM reranking in 5 minutes.
This open-source tool adds a smart reranking layer to your existing Elastic queries using ZeroEntropy's API.

🧠 Built for developers who want great search results without rebuilding their stack.

---

🖥️ Demo UI Preview

Interactive A/B testing interface comparing original Elasticsearch results with ZeroEntropy LLM-reranked results

---

🎯 The Problem

Most Elasticsearch users struggle with:

• Vague queries that return irrelevant results
• Complex search intent that BM25 can't understand
• Ambiguous terms that need contextual understanding
• No AI skills to implement vector search

Solution: Add LLM-powered reranking without changing your existing setup.

---

✨ Features

• 🔌 Plug-and-play: Drop-in proxy or client wrapper
• 🧠 LLM-powered relevance via ZeroEntropy API
• ⚡ Works with any Elasticsearch or OpenSearch backend
• 🔍 Hybrid-ready: Rerank BM25 + vector results
• 📊 Side-by-side comparison: Original vs reranked output
• 🚀 5-minute setup: No AI expertise required
• 💰 Cost-effective: Only rerank top N results

---

🚀 Quick Demo

🖥️ Interactive Demo UI

Try our live A/B testing interface to compare original vs reranked results:

# Quick start (recommended)
./run_demo.sh

# Or manual setup
cd examples
pip install -r requirements_demo.txt
streamlit run demo_ui.py

Features:

• 🔄 Side-by-side comparison
• ⚙️ Interactive parameter controls
• 📊 Real-time performance metrics
• 🎯 Demo mode (no API keys needed)

💻 Code Example

import asyncio
from elastic_zeroentropy import ElasticZeroEntropyReranker

async def main():
    async with ElasticZeroEntropyReranker() as reranker:
        response = await reranker.search(
            query="machine learning applications in healthcare",
            index="research_papers"
        )
        
        for result in response.results:
            print(f"🔍 {result.document.title}")
            print(f"   📊 Score: {result.score:.4f}")
            print(f"   📄 {result.document.text[:100]}...")

asyncio.run(main())

Before vs After:

Query	Original Elastic	Reranked (ZeroEntropy)
"books about hope and war"	🔻 Generic results	✅ Contextually relevant
"machine learning healthcare"	🔻 Broad matches	✅ Domain-specific papers
"python async programming"	🔻 Mixed relevance	✅ Focused tutorials

---

🖥️ Demo UI Guide

The Elastic-ZeroEntropy Demo UI is a powerful A/B testing interface that lets you compare original Elasticsearch results with ZeroEntropy reranked results in real-time.

🚀 Getting Started

1. Launch the Demo:

   ./run_demo.sh
   

   This opens the UI at http://localhost:8501

2. Choose Your Mode:

   • Demo Mode (default): Works without API keys using sample data
   • Live Mode: Connect to your Elasticsearch + ZeroEntropy API

🎛️ UI Features

Main Interface

• 🔍 Search Bar: Enter your query to test
• 🚀 Run A/B Test: Execute the comparison
• 📊 Results Comparison: Side-by-side view of original vs reranked

Sidebar Configuration

🔑 API Configuration:

• ZeroEntropy API Key: Your API key from https://zeroentropy.dev
• Elasticsearch URL: Your ES instance (default: http://localhost:9200)
• Index Name: Target Elasticsearch index

🔍 Search Settings:

• Initial Results from ES: How many docs to retrieve (10-200)
• Documents to Rerank: How many to send for reranking (5-50)
• Final Results: How many to display (5-20)

🧠 Reranking Settings:

• ZeroEntropy Model: Choose zerank-1 or zerank-1-small
• Combine Scores: Mix ES + rerank scores
• Score Weights: Adjust ES vs rerank importance (0.0-1.0)

📊 Demo Data:

• Use Demo Data: Enable for testing without real connections

📈 Understanding Results

Performance Metrics

• Query Time: Total search + reranking duration
• ES Time: Elasticsearch query time
• Reranking Time: ZeroEntropy API processing time
• Model Used: Which ZeroEntropy model was applied

Results Comparison

• Original Ranking: Elasticsearch BM25 results
• Reranked Results: ZeroEntropy LLM-enhanced results
• Score Breakdown: ES score vs rerank score vs combined score
• Relevance Improvement: Visual indicators of better matches

Visual Features

• 📊 Performance Charts: Real-time metrics visualization
• 🎯 Relevance Indicators: Color-coded relevance scores
• 📋 Detailed Metadata: Document info, timestamps, sources
• 🔄 A/B Comparison: Side-by-side result analysis

🎯 Use Cases

Testing Search Quality

1. Enter a complex query like "machine learning healthcare applications"
2. Compare original vs reranked results
3. Notice how LLM reranking improves contextual relevance

Parameter Tuning

1. Adjust top_k_initial to control ES result pool
2. Modify top_k_rerank to balance cost vs quality
3. Tune score weights for your use case
4. Test different ZeroEntropy models

Performance Optimization

1. Monitor query times for different configurations
2. Balance speed vs accuracy with model selection
3. Optimize for your specific workload

🔧 Advanced Features

Real-time Configuration

• Dynamic Parameter Updates: Change settings without restart
• Live API Testing: Test your ZeroEntropy API key
• Elasticsearch Connection: Verify ES connectivity

Demo Mode Benefits

• No API Keys Required: Test the interface immediately
• Sample Data: Realistic documents for testing
• Simulated Reranking: Understand the process flow

Production Mode

• Live Elasticsearch: Connect to your actual ES instance
• Real ZeroEntropy API: Use actual LLM reranking
• Performance Monitoring: Track real-world metrics

💡 Pro Tips

1. Start with Demo Mode: Understand the interface before connecting APIs
2. Test Complex Queries: Try ambiguous or multi-concept searches
3. Compare Different Models: Test zerank-1 vs zerank-1-small
4. Monitor Performance: Watch query times and adjust parameters
5. Save Configurations: Note your best settings for production

🚨 Troubleshooting

Demo won't start:

# Check if port 8501 is free
lsof -i :8501
# Kill existing process if needed
pkill -f streamlit

API connection issues:

• Verify your ZeroEntropy API key
• Check Elasticsearch URL and connectivity
• Ensure index exists and is accessible

Performance problems:

• Reduce top_k_initial for faster queries
• Use zerank-1-small for lower latency
• Adjust score weights for your use case

---

📦 Installation

pip install elastic-zeroentropy

Or with CLI tools:

pip install "elastic-zeroentropy[cli]"

---

⚡ 5-Minute Setup

1. Get ZeroEntropy API Key

# Sign up at https://zeroentropy.ai
export ZEROENTROPY_API_KEY="your_api_key_here"

2. Configure Elasticsearch

export ELASTICSEARCH_URL="http://localhost:9200"

3. Run Your First Search

from elastic_zeroentropy import ElasticZeroEntropyReranker

async with ElasticZeroEntropyReranker() as reranker:
    results = await reranker.search("your query", "your_index")

---

🎛️ Advanced Usage

Custom Reranking Configuration

from elastic_zeroentropy import RerankerConfig

config = RerankerConfig(
    top_k_initial=100,      # Get 100 docs from Elasticsearch
    top_k_rerank=20,        # Send top 20 to reranker
    top_k_final=10,         # Return top 10 final results
    model="zerank-1-small", # Use faster model
    combine_scores=True,    # Mix ES + rerank scores
    score_weights={
        "elasticsearch": 0.3,
        "rerank": 0.7
    }
)

response = await reranker.search(
    query="deep learning neural networks",
    index="ai_papers",
    reranker_config=config
)

Batch Processing

from elastic_zeroentropy import SearchRequest

requests = [
    SearchRequest(query="machine learning", index="papers"),
    SearchRequest(query="natural language processing", index="papers"),
    SearchRequest(query="computer vision", index="papers"),
]

responses = await reranker.search_batch(requests, max_concurrent=3)

Health Monitoring

health = await reranker.health_check()
print(f"Status: {health.status}")
print(f"Elasticsearch: {health.elasticsearch['status']}")
print(f"ZeroEntropy: {health.zeroentropy['status']}")

---

🖥️ Command Line Interface

Basic Search

elastic-zeroentropy search "machine learning" my_index --top-k 5

Advanced Search with Filters

elastic-zeroentropy search \
  "deep learning" \
  research_papers \
  --top-k-initial 100 \
  --top-k-rerank 20 \
  --top-k 10 \
  --model zerank-1-small \
  --filters '{"category": "AI", "year": {"gte": 2020}}' \
  --debug-info

Health Check

elastic-zeroentropy health

---

🏗️ Architecture

User Query → Elasticsearch → Top N Results → ZeroEntropy API → Reranked Results

1. Elasticsearch Search: Get initial results using your existing queries
2. Document Selection: Choose top N documents for reranking
3. ZeroEntropy Reranking: Send to LLM for contextual relevance scoring
4. Score Combination: Mix Elasticsearch and rerank scores (optional)
5. Final Results: Return intelligently reranked results

---

🎯 Performance

Based on ZeroEntropy's benchmarks:

• Accuracy: Up to 28% improvement in NDCG@10
• Speed: ~150ms latency for small payloads
• Cost: $0.025 per million tokens (50% less than competitors)
• Throughput: High concurrent request support

---

🔧 Configuration

Environment Variables

# Required
ZEROENTROPY_API_KEY=your_api_key_here

# Optional
ZEROENTROPY_BASE_URL=https://api.zeroentropy.dev/v1
ZEROENTROPY_MODEL=zerank-1
ELASTICSEARCH_URL=http://localhost:9200
DEFAULT_TOP_K_INITIAL=100
DEFAULT_TOP_K_RERANK=20
DEFAULT_TOP_K_FINAL=10

Programmatic Configuration

from elastic_zeroentropy import ElasticZeroEntropyConfig

config = ElasticZeroEntropyConfig(
    zeroentropy_api_key="your_key",
    elasticsearch_url="http://localhost:9200",
    default_top_k_final=15
)

---

🧩 Coming Soon

• Smart Reranking: Only rerank when needed (query classifier)
• UI Toggle: Compare original vs reranked results ✅
• LangChain Integration: Plugin for LangChain workflows
• Streamlit App: Interactive demo and testing ✅
• More Models: Support for additional ZeroEntropy models
• Caching: Intelligent result caching for repeated queries

---

🤝 Contributing

We're building a better search UX together! PRs, feedback, and ideas welcome.

Development Setup

git clone https://github.com/vagabond11/elastic-zeroentropy.git
cd elastic-zeroentropy
pip install -e ".[dev]"
pytest

Areas to Contribute

• 🎨 UI/UX: Better demos and visualizations
• 🔧 Integrations: LangChain, Streamlit, FastAPI plugins
• 📊 Analytics: Search quality metrics and monitoring
• 🌐 Documentation: Examples, tutorials, case studies

---

📊 Community Stats

---

🏆 Why This Works

Most Elastic users struggle with vague, complex, or ambiguous queries.

This tool reranks the top N results using LLM-based contextual understanding via ZeroEntropy.

No vectors? No problem.
No AI skills needed.

Just drop in the reranker and get smarter search results instantly.

---

📄 License

MIT License - see LICENSE for details.

---

🔗 Links

• 📦 PyPI: https://pypi.org/project/elastic-zeroentropy/
• 🐙 GitHub: https://github.com/vagabond11/elastic-zeroentropy
• 🤖 ZeroEntropy: https://zeroentropy.ai
• 📋 Issues: https://github.com/vagabond11/elastic-zeroentropy/issues
• 💬 Discussions: https://github.com/vagabond11/elastic-zeroentropy/discussions

---

Made with ❤️ by houssam Ait O.

Keywords: elasticsearch, search, ai-search, reranking, llm, zeroentropy, semantic-search, bm25, relevance, hybrid-search, open-source-search