How AI Can Empower Your Current Business Applications

Preface

In today's fast-changing tech world, the buzz about Artificial Intelligence (AI), especially OpenAI and ChatGPT, is hard to miss. The promise of the future is exciting, but there's a challenge: how to actually use these advanced capabilities in your current business applications.

Unlocking the Power of Semantic Search

One of the potential applications that stand out is semantic or "human language" search. Imagine being able to search and retrieve information using natural language queries, much like you'd ask a question to a human. It's an idea that holds incredible promise for streamlining interactions and extracting insights from vast amounts of data. However, understanding how to implement this concept can be a difficult task.

The How: Utilizing Large-Language Models

So, how can we make the most of this potential? Here's where the power of Large-Language Models (LLMs), e.g. from OpenAI, comes into play. By utilizing these models, we can dynamically generate vector representations of documents as they're created or updated. When a search query arises, the LLM Model generates a vector representation too and in the application, using vector similarity search, we can retrieve relevant documents. This approach echoes traditional methods like TF/IDF-based search, but with a crucial difference – we're obtaining numerical representations from LLMs.?

Introducing the Empowering Solution

Recognizing the gap between potential and implementation, I started to develop a Java tool that helps n accomplishing the task. With just a few lines of configuration or declaration, your Java Spring application can already seamlessly integrate a "human language" search service. Here's how it works:

• Once the necessary dependencies are added, proceed to utilize a predefined Entity-Listener for JPA entries that hold free text and appropriately annotate the text-getter method.
• This listener places a message in a message queue when an entity is either updated or created. Subsequently, the processor updates its own vector database with a vector representation of the text attribute, sourced from a LLM (currently OpenAI), as required.
• When it comes to the search service, it acquires a vector representation of the query's text from the same LLM. It then locates the suitable document within the vector database using a vector similarity search.

To illustrate the concept in action, I've prepared a simple demo "recruitment" application hosted on Google Cloud Platform. You can find an overview of this demo on YouTube https://youtu.be/QF1oro2YQaQ. The demo itself is on https://llm-integration-demo.ew.r.appspot.com/gui/candidate-list with a search form https://llm-integration-demo.ew.r.appspot.com/llm-integration/ui/search-one.

The source code for both the project and the demo application is available on GitHub: https://github.com/mxn/llm-integration and https://github.com/mxn/llm-integration-demo correspondingly

Summary

The excitement around AI is well-deserved, yet the real task is leveraging its potential to achieve concrete advantages. Semantic search stands as a shining example of a practical application. With the help of the tool, incorporating a "human language" search can become a simple task, giving way to a more user-friendly and efficient search experience. The choice of LLM is not limited to OpenAI alone; there are various other commercial and open-source options that could be more in line with security and other considerations. Feel free to contact the author if you're interested in implementing such kind of solution.