TechPrufer
Sign inGet started

Blogs›MongoDB›Vector Representation Optimization

Vector Representation Optimization

Vector Representation Optimization

2
posts
2025

Introduced Matroyshka Representation Learning (MRL) to reduce the storage and compute costs of vector search systems. MRL allows for the use of lower-dimensional vector representations that approximate the similarity of full-fidelity representations, enabling a trade-off between cost and accuracy. The post details experiments with Voyage AI's models, demonstrating that using 512-dimensional vectors can achieve comparable retrieval accuracy to higher-dimensional vectors while significantly reducing storage and compute. This post further details the integration of Voyage AI's embedding and reranking models into MongoDB Atlas, enabling developers to leverage these advanced AI capabilities for semantic retrieval. It highlights the benefits of native integration for simplifying developer workflows, enhancing accuracy, performance, and cost efficiency, and discusses future explorations in multi-modal retrieval and instruction-tuned retrieval.

2025

Lower-Cost Vector Retrieval with Voyage AI’s Model Options

8/6/2025

This post introduces Matroyshka Representation Learning (MRL) as a technique to reduce vector dimensionality for cost and performance optimization in vector search. It explains the concept of MRL, demonstrates its application with Voyage AI's models (voyage-3-large, voyage-3.5, voyage-3.5-lite) by showing cosine similarity scores across different dimensions, and presents experimental results (NDCG@10, MRR@10, storage costs) comparing vector search indexes with varying MRL dimensions (256, 512, 1024, 2048). The findings suggest that 512-dimensional vectors offer a good balance of accuracy and cost-efficiency.

Rethinking Information Retrieval in MongoDB with Voyage AI

4/24/2025

This post details the integration of Voyage AI's embedding and reranking models into MongoDB Atlas. It explains how these models enhance semantic retrieval by transforming data into vector representations and improving search accuracy through reranking. The post discusses the benefits of native integration, including simplified developer workflows, improved accuracy, performance, and cost efficiency. It also outlines future development directions such as multi-modal retrieval and instruction-tuned retrieval.