Google Vertex AI Reviews

We are using Google Vertex AI for various purposes, especially various chatbots for customer engineers and customers. Additionally, we are working on projects involving AI and machine learning.

Features like the LLM model are valuable, and we are using Google Vertex AI for these purposes. Additionally, the integration of AutoML features streamlines our machine-learning workflows.

I mostly use LLM models on Vertex AI. When there is a large document or multiple documents, I put them in the index database of Vertex AI's platform and it extracts the right information.

We work with clients who request the implementation of a certain document into a chatbot. Because of the limited knowledge of AI, our task is to link that file to the ML and provide a platform that can work as a customer service. We previously used LangChain Phython, but now it is done through Vertex AI. 

We use it to run AI chatbots, generative AI, and some data science and machine learning instances.

One of the use cases is that we built a chatbot to improve technician efficiency by providing all the user manuals for a specific line of work. The technicians could then interact with the chatbot and get generative AI responses on fixing errors that pop up.

Vertex AI is a playground for data analysts. It is for machine learning engineers and data scientists. We create, test, customize, deploy, and monitor our models on Vertex AI. It is a fully managed product of machine learning.

About twenty people at our company use Vertex AI.

The most valuable feature of Vertex AI is the Feature Store. When it comes to model training and artificial intelligence, a whole team of people was needed to do the work. It could not be done by just one person. Data engineers, machine learning engineers, and data scientists have completely different tasks and, therefore, the communication gap is inevitable. With Feature Store, you can create your features and make them centralized in the store so everyone can see them. The feature is locked and you can create multiple versions of the features. It is similar to GitLab. It has a version control function which is very practical and important for anyone who works in machine learning.

I also like the monitoring feature. I give that feature a ten out of ten. It saves data scientists' lives. After deploying the model and that model runs on the cloud, you need to constantly check on it. With the monitoring feature, you can relax and wait for the notification in case something is wrong. You have time to examine what went wrong. In addition, you can define the error and pre-set the exact thing you want to monitor. Whether it's a task failure, an instant failure, too much uptime, or downtime, the monitoring feature is extremely helpful.

My research was around how well they support RAG use cases. Do they have all the components required for building the RAG use case? As in, you need a data storage part, vector storage where you can store your vector embeddings, and the model layer itself.

Vertex comes with inbuilt integration with GCP for data storage. It also supports ChromaDB, which is not provided by Azure.

We were looking for optimization for a machine learning module. Gurobi is an optimization library, but it's quite expensive. We were looking for an alternative to Gurobi and came across Google Vertex AI. We still find that Gurobi works better than Google Vertex AI. Eventually, we dropped out after six months of using Google Vertex AI.

Google Vertex AI is an out-of-the-box and very easy-to-use solution. You don't need an army of data scientists to help you build your model. The solution is really easy to use. If we teach or tell the business people what to select and how to train, they can do it by themselves in three months. That's how easy it is to use Google Vertex AI.

Our use case involves leveraging it for tasks such as generating document summaries with keyword detection after scanning. Additionally, we employ image augmentation based on image descriptions. For advanced language analytics, we analyze the frequency of specific keywords and business terms within documents, ultimately ranking the documents based on these criteria.

It provides the most valuable external analytics. Whether I'm searching for patterns or conducting keyword-based analytics, I consistently achieve better accuracy with Vertex compared to OpenAI from Azure.

In my company, I use the Google Cloud Platform solution for Vertex AI Workbench and Vertex AI with Gemini. I also use Vertex AI Agent Builder and the other stuff they offer as a framework to build different kinds of applications.

I won't be able to give you the specific details for the customers, but it is mainly for building the operational application for our company's clients by focusing on GenAI capabilities so that at least we can optimize the work, which usually comes in the agent support.

The most valuable features of the solution are that it is quite flexible, and some of the services are almost low-code, with no-code services, so it gives agents flexibility to build the use cases according to the operational needs. If somebody wants to build something on their own, it is very easy for even the end users. On the contrary, if they want to customize the pipelines and cover the granular details, they can utilize the options available for capturing and controlling every pipeline stage with codes.

The tools' models are pretty good. The multi-model capabilities from Vertex AI with Gemini 1.5 Pro are quite good, but the cost is slightly higher.