Amazon Bedrock Reviews

The principal use case for Amazon Bedrock that we are working on is regarding a logistic company that has a flow where they receive emails to look for incoming invoices. We have an architecture that sends these invoices to the queue and identifies the partners involved on the invoice, and we have specific queues; for each queue, we have a specific agent that will treat this information and analyze what they are looking for in the invoice they are sending.

For example, if we have to send it to the invoice team, we read this email, and based on the request in the email, we get the information from the email body and then the information regarding the invoice. We understand the behavior we have to deliver to the specific system or how we can create a new ticket for the invoice team to get this invoice and register it on the legacy system. It involves this integration and using specific agents to read, understand, and process the incoming invoices.

One of the best features of Amazon Bedrock is that it is easy to use, and users do not have to worry about the infrastructure. For example, if we have to create a similar solution, we have to think about what kind of LLMs we need to create for it to execute. Using the Amazon Bedrock architecture, we choose some of the available LLMs and then can focus on the business solution.

The AI solution, this specific intelligent agent that we are creating, is not something that you can just execute; you have to create some logic around the agents to create this process, and then they will execute what you have created for them. Amazon Bedrock accelerates the process of creating your architecture. It is not specific to modernize your legacy, but you can accelerate it.

They provide all the infrastructure that you need; you just have to follow the specifications, and then we are able to create our agents to solve the kinds of problems mentioned earlier. For content creation capabilities, they help us streamline our content generation processes across different departments. This solution we are talking about is just to process one specific process that we are looking for, which is totally inside the same area. We are not creating it for or interacting with other areas inside the company. However, once you have your infrastructure, even if it is hybrid and not only on AWS, you can integrate it.

Today, the solution we are specifying is a kind of LEGO that you can just get some pieces and connect in our solution. For example, we are using SNS to send these queues. You just need to connect an outside AWS solution and send it through API to this queue; you put the message there, and you can get it inside our process. Even if this email is not the specific trigger that will call our solution, we can get it from SAP. If I have this same invoice and want to use this process, you just have to send it through an API; we have just connected it to our solution, and you can create it to be available for the other areas or other systems involved in the company legacy systems.

I have to gain more maturity to provide some improvements to Amazon Bedrock. I have a lot to do with the environment they already provided. For example, they are able to connect to any LLM solution such as Llama, Meta, Gemini, or ChatGPT. It is open; you just choose your favorite LLM solution, and you can integrate it into Amazon Bedrock. We have a lot of possibilities to do this integration at this moment; we just need to work on it, create more maturity, and then we can provide some enhancements that we can see on the solution as a whole.

For companies in general, the main pain point or main issue related to Amazon Bedrock is security because they are not confident that all information is hidden by this kind of architecture. They wonder if they are providing some company information that can run away, and I think that is the challenge we have now. We need to find ways to work on it and make our clients' data secure. They are looking for that to guarantee that this is a great solution for companies that is also secure.

I would rate Amazon's support an 8 or 9 because we have a lot of information on the portal, and we can do a lot of things independently. The solutions we have with AI also help us work on it; it is easier than ever before because the AI integration works with us on a daily basis. For example, I already know this architecture; I just need to create something specific for this job. I just send it to an AI solution Claude or even Cursor, which is very good for developers, and then provide the logic and code; you just have to adopt something, it is a little customization, and then you have it ready. We are experiencing the fastest time ever to get things done with AI integrating into our work, regardless of where we are.

Positive

For data analysis, it depends on the client's decision. For this specific case, we are working only on this solution because the data analysis for this company's corporate solution is done with Power BI, not Amazon Bedrock.

The impact of Amazon Bedrock's sophisticated natural language processing on our company's ability to predict future outcomes is very interesting because, before we were using some Python codes, we created server instances to upload it, and we had some difficulty integrating it with the ecosystem because all the features we were creating were manually based. If I want to create a specific agent to connect to the client architecture, I have to do it manually. Once we use Amazon Bedrock as a composed architecture, it is easier to not only connect and provide the infrastructure as a service, but you can easily deploy it in the production environment because if you are on your machine or laptop, you can run these Python codes to do what we are testing or trying to solve, but it is not scalable.

The integration with other AWS services contributes to cost savings. For example, you can use a Lambda to create that specific solution; the solution we create to read the emails is from Lambda, and we connect through the Outlook component to get this message and send it in a JSON file to the SNS queue. I do not have to create any server or anything else related; I just create the Lambda service connected to the SNS services, and this integrating environment is all serverless.

Regarding Amazon Bedrock's pricing, for this specific case with our client, before they had a machine that worked for about $500 per month, and once we evolved it to this new architecture, they paid around $2,000 for the same solution. If you compare it only with the Python code we were running before, it is three times the price, but once you have it on scale, you can share it with other solutions. Once you decide to use this in a corporate way that will scale with other areas and have a well-defined architecture for your company, you can share it, and it becomes a fair price to pay for this kind of right solution.

I rate Amazon Bedrock an 8 out of 10.

I am currently working on Amazon Bedrock Agent Core. We have created a data pipeline where we are using Amazon Bedrock Agent Core primarily for transformation. We use the agent for custom rules, transformation, and data quality checks. We are trying to create an Agentic AI data pattern using Agentic AI, where the agent will take decisions.

We are using the SDK framework of AWS. We have used Amazon Bedrock for creating custom rules by passing prompts to the agent and to an LLM model. The entire project is deployed on AWS Cloud, and currently we are doing a POC with an account provided by AWS.

Amazon Bedrock Agent is what we are currently using. It is easy to connect to LLM, and the integration is very simple and straightforward.

Amazon Bedrock is very easy to configure and launch. It takes care of the creation of Docker images with CodeBuild and putting the images into the ECR repository. All of these things happen very simply and easily with the help of Agent Core and Amazon Bedrock Agent Core.

Overall, I find Amazon Bedrock simple and good, and I would give it a rating of seven or eight.

Currently, I do not have any negative points in mind about Amazon Bedrock because I think Amazon Bedrock and other services are good. We have to use OpenSearch as well. We have not implemented RAG in our data pipeline, but I have studied OpenSearch and I think it is good. I do not see any negative points so far.

The end-to-end application setup integration was very difficult. An end user needs a URL where a page should appear and they should be able to use it. We created one ECR inside one VPC and created one EC2 instance inside that VPC, and we copied our data into the EC2 instance. From there, we launched Agent Core and ran it. In that case, we had to give the public URL to the user, which was not secure. The integration part was very difficult.

I have been working with Amazon Bedrock for three months.

Stability is good. I can give it ten out of ten, but I would say eight out of ten.

I have not tried to scale up the Glue things, but I think it is possible. You can scale up the number of DPUs or nodes dynamically based on your workload using the agent, but we have not tried that. The challenge for me was the integration.

I have not taken any support from Amazon, so I cannot rate this right now. Once we asked for support, but it was not very good. I will not say the support from Amazon was bad because they had given some training. We asked them, but it was not that good. Perhaps we were not clear about the requirement. I will not blame them entirely, but my improvement idea is training videos.

I have used Google Vertex AI in the past. Since this project is on AWS and in the AWS environment, I get the SDK for this one. Instead of using LangGraph, I can go for the AWS framework and Agentic framework.

I found Google Vertex AI also very good. We are using AWS because of the client requirement, as they have their setup in the AWS environment.

The setup is simple, but when it comes to an entire end-to-end application, an end user needs a URL where a page should appear and they should be able to use it. The integration part for this was very difficult.

For operational data engineering, this can definitely be applied. We are currently doing a POC, and it is not in production right now.

When it comes to AWS training material, I find that compared to GCP, GCP has very good demo-style training material with videos from start to end and workshops. Their demos are like workshops with a sandbox, instructions, and video. AWS has material in readable format where I have to read the document. I would prefer a demo of thirty minutes or one hour where I can implement something end-to-end so that I get an idea to implement in my project as well. It may not be exactly the same, but it would cover the end-to-end part.

I will recommend Amazon Bedrock to other users because of the Strand SDK. I am not sure what Google has with them, but it is easy. If I use LangGraph or any other framework, they change because they are open source, so if they bring a new version, we may not know about it. I think Strand SDK is good. It is like LangGraph only. I will recommend it, definitely.

I would rate my overall experience with Amazon Bedrock as a seven out of ten.

Amazon Bedrock is one of the services which is similar to ChatGPT. From my perspective, I was creating a version of ChatGPT that would answer my customers' questions. The significant difference between them is that ChatGPT provides answers to all questions, whereas the Bedrock instance I created was specifically designed to answer questions related to a particular business. My website is a wine selling application, and it was configured to answer only wine-related questions, such as health benefits, consumption guidelines, and product recommendations.

The customization is important for my use cases because unlike other GPT or APIs that can give answers to all questions regardless of the topic, Amazon Bedrock answers questions based on the knowledge base we attach to it. I usually provide a PDF file with knowledge in it. After that, the vectors are created, and Amazon Bedrock is able to give answers to particular questions. If it does not have the knowledge about the question, it will give a fallback intent. For instance, I added a restriction rule that if someone asks about the disadvantage of a particular thing, it should stop that question and fallback. When someone asked how wine could be used for entertainment purposes, that question was blocked by the system.

The most valuable features of Amazon Bedrock include scalability, ease of access, having 20 to 25 plus pre-trained models, and scaling capabilities. It can adjust its scalability as needed.

The valuable features that have helped in leveraging generative AI for operational efficiency improvements include customization capabilities, various types of models suitable for specific use cases, and the integration of knowledge bases. It has the capability to convert normal documents to vectorized documents, and can provide answers based on these documents.

The only aspect that could be improved about Amazon Bedrock is the pricing. There are no additional features or improvements needed for the product itself.

I have been working with Amazon Bedrock for six months as part of my project.

The stability of Amazon Bedrock is good as I have not faced any issues. The only consideration is that during model selection, we need to select the proper model for our use case.

Regarding scalability, Amazon Bedrock has the capability to answer all questions and handle traffic as needed. The only consideration is pricing, as it increases with traffic since charges are based on prompt requests.

I have had interactions with the technical support of Amazon Bedrock. My experience with the technical support has been very good because they resolved my billing issue within a day.

Positive

The initial setup of Amazon Bedrock is somewhat complex as it requires integration with two to three services. We must set up the console cloud and provide knowledge to the system. Subsequently, we need to create an AI model to generate answers through Amazon Bedrock. Then, we must integrate that AI model into our code or UI, and create roles and restrictions. This makes the process complicated.

For content creation, it is necessary to use another model for that purpose.

Amazon Bedrock can be integrated with other services such as Amazon Lex. It can also be integrated with knowledge bases, and these integrations can be incorporated into code and deployed to the EC2 server or any serverless platform.

The main benefits derived from Amazon Bedrock are its customization capabilities, model variety, and the ability to integrate into UI systems.

On a scale of 1-10, I rate Amazon Bedrock an 8.

We have been mainly using Amazon Bedrock for LLM invocation. We explored it for RAG pipelines as well. We did not do extensive exploration for agentic AI. When we started, it was not very developed at that time. There may be some progress now, but we progressed with our solution so it was not really a need after starting on an implemented solution. Our main focus has been on the invocation of the LLM models and the tracking of logging and cost tracking.

For analysis or report generation, we are using Amazon Bedrock mainly. We are doing some processing on the data before providing it for the use case in order to generate specific content. There is a processing step before providing the data. It is not provided in raw format.

The feature that is most valuable to me and our organization is enabling the use of large language models through API calls. The invocation of the models is the feature that has helped a lot, enabling user models, mainly closed models.

The enabling of the call or the invocation of the models through API calls is the main feature. The following features were not, at least from my perspective, as valuable as that. I did not use fine-tuning on Amazon Bedrock. It can be a good feature, but the one that helps the most is calling the LLM or invoking the LLM models.

I think the impact is provided or enabled through the large language models. They are NLP models. They enabled the generation or the creation of information in textual form. This is coming from the capability provided by large language models.

I may not be the right person to provide this insight or recommendation. I am not really following the progress in it. Some new features may have been implemented and I am not aware of it. I would have to go back to Amazon Bedrock for that information. Working on agentic tools and providing more capability on agentic AI is what was missing at the last time I explored Amazon Bedrock.

As I recall now, I found some limitations or some models are not present in Amazon Bedrock sometimes while they are present in other platforms. There are new models that are coming, and sometimes I cannot find them in Amazon Bedrock. The limitation of region sometimes also caused some problems. Some models are found in regions and not in others, and the customers or the clients do not want the model to be hosted in a region different than their own. These are some limitations that we encounter sometimes while working on a product for a specific customer.

I have been using Amazon Bedrock since 2023.

I used OpenSearch and SingleStore. I mainly used OpenSearch, in fact. I think that time I was doing some comparison between them. It was a light solution and of course it was open source so it was free to use. I used this in an application, I think it was a RAG application that time, but it has been a while since then.

When I saw the comparison, even without using Pinecone mainly, Pinecone has a lot of advantages but the main limitation was that it was not free, so I opted towards a free solution, an open-source and a free solution. I have experience with Elasticsearch, so that is why it was easy for me to use OpenSearch. Overall, I think it was a good experience. It was easy to use and lightweight. That is what I remember mainly.

Amazon Bedrock enabled the use of huge models and the democratization of their use at comparatively low cost, if we host these models in the company. The capability to switch from one model to another is an important advantage.

The price of invoking the model is considerably better compared to hosting the model with our local resources. This is an advantage for Amazon Bedrock.

The other option might be testing on-premise hosting of the models, especially for small language models. We are still in research, so this may be an option in the future for agentic AI.

Since that time, I have worked on agentic AI, using LangGraph and LangChain, mainly tools for agentic AI. I also used LangSmith. I have experimented with many tools and frameworks related to agentic AI, including Crew AI. There is a not very well-known framework for agentic AI which is Agemo. I found some papers or the website of this framework cited that they are very fast. This framework was agentic AI focused.

I do not recall using some of these other tools. I am using Amazon Bedrock.

I would rate this product an 8 out of 10.

Amazon Bedrock is designed for LLMs (Large Language Models). These are not generally things you can host easily on SageMaker. You have to build up custom environments for that, particularly if you want access to the proprietary models. You cannot access the proprietary models such as Claude on SageMaker; you have to use Amazon Bedrock.

With all of the Amazon products, including Amazon Bedrock, I can say that Bedrock has only been in use since it was launched about 18 months ago. Other areas in our business use other cloud services.

Amazon Bedrock's main value lies in its LLM capabilities. These are not generally things you can host easily on SageMaker. You have to build up custom environments, particularly if you want access to the proprietary models. You cannot access the proprietary models such as Claude on SageMaker; you must use Amazon Bedrock.

What could be improved for Amazon Bedrock to make it more mature is that AWS needs to consider bringing their platforms together, and not having different ML and AI platforms.

We started using Amazon Bedrock 18 months ago when they launched it.

The technical support specifically for Amazon Bedrock is acceptable, but this space is very new. It depends on who you talk to. These are all new platforms, and scaling or even building AI solutions and scaling ops is a completely new concept. Most vendors do not know how to do it just yet. We work with them, but we have to build a lot of components ourselves.

The technical support specifically for Amazon Bedrock is acceptable, but this space is very new. It depends on who you talk to. These are all new platforms, and scaling or even building AI solutions and scaling ops is a completely new concept. Most vendors do not know how to do it just yet. We work with them, but we have to build a lot of components ourselves.

Neutral

If you look at a tool such as Databricks, all of that functionality is integrated in one platform. I do not have to worry about stitching together a whole bunch of services to make it happen.

The problem is both with Amazon themselves and with my complex environment. Their tools are not as simple or intuitive as some others, but our environment is also complex. It is a multi-cloud, multi-region setup.

The most problematic integrations are not so much problematic; it is the fact that our data is scattered all over the world in different clouds and different regions. We have to consider networking and other security aspects to manage this.

On a scale of 1-10, I rate Amazon Bedrock a 6 out of 10.

We have a fairly extensive use case library that includes everything from document and report generation on the simpler side to full project management on either the hardware or software side. 

We build use cases to enhance processes within standard business practices, like automating call centers, customer experience, or customer success functions. We work on financial reporting within that side as well. 

The AI platform and supportive machine learning models are designed to rapidly prototype any kind of enterprise business use case. Once we prototype a use case, the platform becomes increasingly smarter, faster, and stronger, making that use case more robust and helping it deliver greater cost savings and efficiency. 

We cannot solve every problem, however, we have a fairly extensible capability. The advantage of Bedrock is not as an amazing enabler of AI platforms, yet we utilize it to deploy application services and microservices within Bedrock ecosystem and leverage prequalified foundation models like Claude and others.

They provide a security umbrella for us. Our encryption keys are managed through Bedrock ecosystem by our clients. Data encryption while in transit and at rest is managed through Bedrock account. Amazon constantly adds new foundation models to Bedrock. When a new model is added, we adopt it, creating a roadmap for our Amazon clients. Secure operation in Bedrock means secure operation in EC2 and S3 buckets, allowing us to leverage processing power and NVIDIA chipsets to improve performance. This is valuable to our application and service layers.

I am not going to speak to their roadmap. Amazon operates Bedrock as an ecosystem supporting third-party models. I am speculating here, but I am sure those third-party models will always be present. 

However, one must consider that Amazon native models could proliferate Bedrock in the future. We would welcome Amazon native models to Bedrock, since, if they are natively built by Amazon, they are tuned to SageMaker and other Amazon service layers. 

They have done this somewhat for generative AI, however, in AgenTek AI business, the only foundation models we can rely on for scaling now are the Cloud 3.5 models like Haiku and SONNET, designed for low latency and complex AI business use cases.

I have used the solution for about five years now.

It can get as big as needed. I hope we explore that option. However, it is scalable on a truly global basis.

With any technical support, it is important to remember that Amazon excels at answering questions related to their ecosystem and architecture. They cannot be expected to be experts on your application and service. So, you always have to bridge the gap by presenting scenarios, getting recommendations, and testing or validating those assumptions. It can get incredibly complicated regarding how much load you put into S3 or EC2. Nonetheless, the documentation and support are generally very good.

Positive

AWS expertise within the team took us three to four months to solve all the problems. Without preexisting expertise, it is uncertain how long it might take. It is less about the time to deploy Bedrock and more about the skills within your team. There might be skill set deficiencies required to make the application or service enterprise capable. Software companies serving SMBs might lack enterprise scaling assets. 

Overcoming hurdles in data encryption and integrity topics is not easy and requires extensive Python programming.

The solution is based on utilization and follows a consumption model. If no load is put into Bedrock, there would be no cost. Processing large volumes of data by the hour incurs high costs.

You are paying for CPU cycles and memory, so how well you design your application to take advantage of these is crucial. A well-designed application requires limited resources and has low costs, while one reliant on Amazon's processing power and memory could experience high costs. 

Our cost is incredibly low, operating for a few hundred dollars a month in production.

It is the best solution in this category and is rated a nine out of ten. 

There is always room for improvement, however, it is a world-class ecosystem.

We are using Amazon Bedrock for OCR solution. It is for manufacturing customers who want their invoice processing to process their PDF documents to get the unstructured data and put it into a database.

We are using the AI capability of Amazon Bedrock to extract data from PDFs.

We work with Snowflake sometimes, Databricks sometimes, and use the AWS platform. I am working on all three.

It was absolutely useful and we found that we are getting 90 to 95% plus success rate while extracting the data from unstructured documents.

It is simple to integrate.

Regarding Snowflake functionality, Snowflake has brought many Cortex features which a simple business user can utilize without having to depend on any technical resource. To use Amazon Bedrock, we need to have significant dependency on technical resources. You need substantial technical capability. The integrations are pretty much simplified in Snowflake, which is not the case in AWS. These are very important and simple things. Now SQL features are being introduced, such as AI SQL in Snowflake.

With the new features introduced, Snowflake is much more user-friendly. The user-friendliness when you give the entire system into the hands of business users is much more powerful than giving these powers to technical users. Business users can build their own solutions and bring their own use cases because they understand the business. With minimal help, they can build their entire solution.

The solutions are getting more complex. For example, regarding Snowflake functionality, Snowflake has brought many Cortex features which a simple business user can utilize without having to depend on any technical resource. However, to use Amazon Bedrock, we need significant dependency on technical resources and technical capability.

We have been using the product for almost two years, though not continuously.

It was quite stable.

Previously, I was using Redshift, Lambda, Kinesis, SNS, and S3. I was using all these services.

We are not using it in much detail. It was one of the use cases, so we used it for that specific purpose.

Sometimes, even now, we are giving solutions for different customers. I am working more in pre-sales plus solutioning.

I am still working with Snowflake and Databricks.

We are using AWS for AI/ML solutions such as Amazon Bedrock for OCR solution.

The implementation was not complex. It was acceptable.

They are absolutely great at any point of time. There is no doubt about that. However, their product is not as mature as Snowflake or Databricks.

On a scale of 1-10, I rate Amazon Bedrock an 8.

I work with an AWS partner, and we offer cloud managed services to our clients as well as reselling services. I've worked with Amazon Bedrock to create solutions, including an image generation solution and a chatbot for an ERP application for schools.

The ability to make changes in the foundational model is valuable since different customers have specific needs, allowing customization. 

Additionally, the no-code application of the service is beneficial since it allows creating solutions without extensive coding knowledge.

The user interface of Amazon Bedrock on the management console needs improvements. It's very bland at the moment.

I haven't worked with Amazon Bedrock for a long time. It took approximately one or two weeks to create a simple image generation solution.

I haven't had any complaints from any customers, so I rate the stability as ten out of ten.

I rate the scalability as an eight out of ten, however, I am not able to provide specific reasons.

We did not require any support from Amazon, so I rate it ten out of ten since there were no complaints.

Positive

The initial setup was rated six out of ten due to some challenges faced by the presales team during the setup.

The price of Amazon Bedrock is a little costly, rated six or seven out of ten. One customer paid around $100 to $200 per month, which was significant given their overall infrastructure costs.

You should be well-versed in AI ML to use Bedrock properly. 

Overall, I rate Amazon Bedrock ten out of ten.

The primary use case of Bedrock is for creating large language models and deploying pretrained models. It is used for generating AI-driven solutions like text summarization, image creation, or text creation. It also incorporates a knowledge base allowing companies to use their own data to gain insights.

Bedrock helps by providing more clarity and improving the efficiency of tasks through quicker solutions to problems. The scalability aspect benefits my project because I can choose different models based on their specialties and because it has connections to other AWS services. This integration allows seamless operation within the AWS platform.

The most beneficial aspect of Bedrock is its pool of models to choose from, catering to specific needs. Additionally, the knowledge base allows companies to connect their data and gain insights tailored to their specific requirements.

Bedrock could be improved by having an API that allows for easy integration with services outside of Bedrock.

I have been working with Bedrock for close to a year now.

Bedrock is stable, and I have found it to be so in my experience.

Bedrock is scalable, allowing the use of different models based on specific needs. The scalability benefits my project through the ability to integrate with different AWS services without leaving the platform.

I have not interacted with customer service or support as I have not had any need for it.

Positive

I have used Gemini and Llama as an API previously.

The initial setup was straightforward and took about five minutes to complete.

Since using Bedrock, benefits include time and resource savings, as it is not expensive and offers similar functionality as individual models based on requests and tokens.

The pricing and licensing cost is fairly reasonable. It is not expensive.

I have experience working with AI infrastructure solutions from Google and Azure, namely Gemini and Llama as an API.

Bedrock is a comprehensive solution that provides everything needed on one platform. It allows for flexibility and scalability, letting users explore different options without needing external resources.

I'd rate the solution eight out of ten.