Grab experimented with combining vector similarity search and LLMs to improve search result relevance. The approach uses vector similarity search (using FAISS and OpenAI embeddings) for initial candidate retrieval, followed by LLM-based reranking of results using GPT-4. Testing on synthetic datasets showed superior performance for complex queries involving constraints and negations compared to traditional vector search alone, though with comparable results for simpler queries.
# LLM-Assisted Vector Similarity Search at Grab
## Overview
Grab, Southeast Asia's leading superapp platform, implemented an innovative approach combining vector similarity search with Large Language Models (LLMs) to enhance search accuracy and relevance. The case study demonstrates a practical implementation of hybrid search architecture that leverages both traditional vector search capabilities and advanced LLM understanding.
## Technical Implementation
### Architecture
The solution implements a two-stage search process:
- First Stage - Vector Similarity Search
- Second Stage - LLM Reranking
### Experimental Setup
- Test Datasets
- Technology Stack
## Production Deployment and Results
### Performance Analysis
- Query Complexity Handling
### Real-World Implementation
- Production Scale
### Challenges and Considerations
- Performance Implications
- Scalability Factors
## Technical Optimizations and Best Practices
### Search Pipeline Configuration
- Vector Search Optimization
- LLM Integration
### Future Improvements
- Areas for Enhancement
## Conclusions and Learning
### Key Takeaways
- Hybrid Approach Benefits
- Production Considerations
### Success Metrics
- Search Result Quality
- System Performance
### Best Practices Identified
- Implementation Strategy
- Production Deployment
Start your new ML Project today with ZenML Pro
Join 1,000s of members already deploying models with ZenML.
