Aeca Blog

We propose a new approach to LLM usage by momentarily reconstructing the context.

Explains the limitations and characteristics of vector embeddings and covers the improvements made to store them.

On June 21, 2024, OpenAI announced the acquisition of database startup Rockset. According to OpenAI, the background of the Rockset acquisition is to improve search infrastructure to make AI more useful. Specifically, what advantages led OpenAI to acquire Rockset?

Explains how to build a natural language search service by applying vector search to a case law search demo using FTS.

Explains the process of downloading case law data and building a case law search service in just one day using Aeca.

We explain the process of data collection and processing, search, and service development for product search using Aeca. Learn how to index when structured and unstructured data are mixed, and how to transform queries for search using LLM.

Starting from March 12 (Tuesday), the product name and company name will change from 'Cognica' to 'Aeca'. The name change will be applied gradually.

You can easily create RAG (Retrieval Augmented Generation) with just one AI database without complex infrastructure setup.

Methods to overcome the limitations of Large Language Models (LLMs) by utilizing Vector Databases (VectorDBs) are gaining attention. To provide accurate answers on specialized information such as law firm case precedents or company communication records—domain data that is not included in the training data—we can use a Vector Database that can convert, store, and search all kinds of data into vector embeddings, serving as a long-term memory storage for LLMs. To illustrate this, we examine a concrete case of how a vector database can complement an LLM through processes like data preprocessing, vectorization, storage, and search, using a Q&A system based on Wikipedia.

Many services are introducing recommendation systems to increase user retention time in modern applications, and this is an important factor directly related to sales, especially in content and e-commerce sectors. Recommendation systems analyze user behavior to understand their interests and provide related items, thereby increasing retention time and inducing purchases. How can vector databases be utilized in this context?

The mobile applications and web services we use have search functions. Most are developed using basic text search provided by databases or full-text search provided by search engines like Elasticsearch. Full-Text Search is one of the traditional methods mainly used for searching text data, focusing on finding specific keywords, words, phrases, etc., in documents, web pages, databases, and more. It typically involves inputting keywords or short sentences to search text data and finding documents that match the keywords, but it does not consider context or semantic similarity.

Vector embedding is a concept that converts various forms of data (documents, images, audio, video, etc.) into arrays of numbers to measure similarity. For example, colors can be represented as three-dimensional vector data in RGB format. By calculating the distance between these vector embeddings, we can determine the similarity between data. This plays an important role in natural language processing, recommendation algorithms, and more. Various data can be converted into vectors through Transformer models, allowing us to measure the similarity between different types of data. For instance, it is possible to measure the similarity between the text "cat" and a picture of a cat in vector space.

Search in web and mobile applications is a core function that shapes a positive user experience. Particularly in commerce services, search goes beyond enhancing user experience and directly impacts company revenue. With the explosive growth of product information and content, search quality—providing timely information that matches the keywords entered by customers—has become a critical factor determining the success or failure of applications and websites. Generally, customers searching for products in commerce services are considered strong potential buyers with a high willingness to pay. It's observed that all actions users take when searching and reacting to search results reflect their purchase intentions, needs, and willingness to spend. Statistically, the conversion rate of users who perform searches is more than twice that of those who don't. Although only less than 20% of the total MAU use search, it's known that over half of the revenue comes from users who have performed a search at least once. Additionally, the churn rate is high for users who fail on their first search, and the conversion rate for customers who re-search is very low. In other words, search is not only a powerful tool to open customers' wallets but also a factor that greatly impacts the sustainability of the service. So, how does search specifically contribute to customer retention, revenue growth, and service improvement?

Search functionality is essential in modern applications like Coupang, Baemin, and Yanolja, where users search for various items. However, developing search functionality is not straightforward. Especially when users expect Google-level search quality, the technical requirements become complex. There are two primary approaches to developing search functionality: using database query capabilities and using a separate search engine. Database query capabilities are suitable for simple searches, but they have limitations for complex search requirements. On the other hand, using a search engine can provide high-quality search functionality but increases development complexity and maintenance challenges. Therefore, early-stage services need to consider how to effectively implement high-quality search functionality.

Search in web and mobile applications is a core function that shapes a positive user experience. Particularly in commerce services, search goes beyond enhancing user experience and directly impacts company revenue. With the explosive growth of product information and content, search quality—providing timely information that matches the keywords entered by customers—has become a critical factor determining the success or failure of applications and websites. Generally, customers searching for products in commerce services are considered strong potential buyers with a high willingness to pay. It's observed that all actions users take when searching and reacting to search results reflect their purchase intentions, needs, and willingness to spend. Statistically, the conversion rate of users who perform searches is more than twice that of those who don't. Although only less than 20% of the total MAU use search, it's known that over half of the revenue comes from users who have performed a search at least once. Additionally, the churn rate is high for users who fail on their first search, and the conversion rate for customers who re-search is very low. In other words, search is not only a powerful tool to open customers' wallets but also a factor that greatly impacts the sustainability of the service. So, how does search specifically contribute to customer retention, revenue growth, and service improvement?

We know that terms like big data, data lakes, and web-scale are fancy and attractive, but those are only everyday issues for very few of us. Most companies will never deal with the petabytes scale of the data. Let's be practical and stay on the ground. Most companies just need a simple but powerful database system to solve real problems. We are here to build a product for most companies, not just for unicorns. Our mission is to solve the common problems often associated with existing database systems and simplify software development by keeping your software stacks as simple as possible.