Welcome to the Green Vectors API
The Green Vectors API is designed to optimize your vector-based workflows by providing advanced tools for efficient vector processing. Whether you need to fine-tune an existing vector or process large batches, our solution is tailored to deliver exceptional performance and scalability. With our API, you can achieve:
Faster Vector Search: Leverage our optimized embeddings to accelerate vector searches, reducing latency and enhancing real-time applications.
Increased Search Accuracy: Improve the precision of your search results by utilizing our algorithm to refine vector embeddings, ensuring more relevant and reliable outputs.
Reduced Storage Space: Benefit from compact, efficient embeddings that minimize storage requirements without compromising quality, lowering operational costs.
How do we achieve these 3 improvements?
Standard vectorization algorithms store vectors as individual rows in a database. Over time, as the number of vectors grows, this structure can lead to significantly larger databases, causing slower search performance and increased storage costs.
Our solution introduces an innovative algorithm that transforms existing vectors into "faceted" vectors. A facet is simply a consolidated representation of vectorized data based on its categorization or organization.
In traditional vector databases, each vector is often associated with a category or description. As new vectors are added, these categories are frequently duplicated, leading to redundancy and inefficiency. For large-scale systems, this repetition contributes to performance degradation and storage bloat.
Our algorithm addresses this challenge by combining all vectors within the same category into a single, optimized faceted vector. This drastically reduces the total number of vectors stored, improving search speed and lowering storage requirements, while maintaining the accuracy and integrity of your data. With this approach, your database remains lean, fast, and cost-effective as it scales.
Example: Standard Vector Database vs. Faceted Vector Database
Scenario: Product Recommendations System
A database stores vector representations of products, each associated with a category, such as "electronics," "furniture," or "clothing."
Standard Vector Database Storage
In most traditional vector databases, every product has its own vector, even if multiple products belong to the same category. Over time, this leads to redundancy as categories are repeated.
Example:
1
Electronics
[0.12, 0.85, ............, 0.23, 0.91]
2
Electronics
[0.45, 0.77, ..........., 0.31, 0.82]
3
Furniture
[0.67, 0.12, ..........., 0.44, 0.99]
4
Electronics
[0.18, 0.89, ..........., 0.34, 0.75]
5
Clothing
[0.56, 0.42, .........., 0.23, 0.68]
Storage Impact: The same category ("Electronics") appears multiple times with individual vectors, increasing storage usage.
Search Complexity: Searching for "Electronics" involves scanning through multiple vectors, which slows down retrieval as the database grows.
Faceted Vector Database Storage
Using our algorithm, vectors within the same category are combined into a single faceted vector. This eliminates redundancy while preserving the ability to search efficiently.
Example:
1
Electronics
[0.25, 0.84, ..........., 0.29, 0.82]
2
Furniture
[0.67, 0.12, ..........., 0.44, 0.99]
3
Clothing
[0.56, 0.42, .........., 0.23, 0.68]
Storage Impact: Only one vector is stored per category, significantly reducing storage space requirements.
Search Efficiency: When searching for results there will be less data, resulting in faster and more efficient queries.
Key Benefits
Reduced Redundancy: Categories are consolidated into a single vector, minimizing storage usage.
Improved Search Speed: Faceted vectors streamline searches by reducing the number of entries to scan.
Scalability: As your database grows, the performance impact is minimized, ensuring consistent and efficient operation.
This example demonstrates how our faceted vector approach simplifies storage and accelerates search performance compared to traditional vectorization methods.
Last updated
