GR8 Tech x AWS AI Chatbot Successfully Closed 50% of Requests

Then It All Started

Imagine you need information about a project and start searching several databases or looking for somebody who can help you find it. This is a classic situation when businesses grow and their databases become more complex: sometimes, operational queries take just minutes, while other times, they require hours of searching.

This is how the idea of an AI chatbot was born—to help employees and clients save time. AI chatbots reduce the time spent searching and improve accuracy. For clients, this means quicker responses and more relevant information. For employees, it is about seamless workflow, helping them focus more on strategic tasks rather than operational ones.

The Problem

To tackle this issue effectively, we engaged with our colleagues at Amazon, who offered their boxed solution.

However, it didn’t fully address our needs. We needed more data customization and a more precise security layer. Finally, our bot is supposed to integrate seamlessly with our databases, including Confluence, Knowledge Hub, and Jira. 

Also, the AWS solution did not provide an option to collect feedback for further advancements and references that the bot used to answer. Additionally, it was essential to make our chatbot easy for our clients to use.

The Solution

To customize the AWS solution, we decided to replace the Confluence data source with the S3 data source. We also filled the S3 repository with documents and developed a custom application based on the Slack_bot framework to replace AWS Chatbots.

In collaboration with AWS, we developed a streamlined solution utilizing AWS Chatbots, AWS Bedrock Agents, and a Bedrock Knowledge Base. We integrated multiple Confluence spaces with the chatbot to assess the project's potential. This implementation was effective from the outset, and we achieved successful results within a few days. Additionally, the source for the Knowledge Base can encompass any content that can be uploaded to Amazon S3.

Based on the evaluation results, we recognized the need for enhancements. Specifically, we required the capability for private messaging concerning sensitive topics, and there was a universal request for source links. Additionally, we needed a mechanism for collecting feedback. Consequently, we opted for a more customizable solution: the native Slack App integrated with the slack_bolt library.

The integration process was notably straightforward. The library's user-friendly interface, alongside the simplicity of the Bedrock SDK, facilitated a smooth implementation. Although employing Python's asyncio presented some challenges, the utilization of libraries such as aiobotocore and aioboto3 proved beneficial. 

This approach effectively addressed most of our concerns; the bot now accommodates private messaging and enables the inclusion of source links. Furthermore, we implemented a security layer that restricts access to certain Confluence spaces for specified users. Ultimately, we piloted the bot and started collecting usage metrics.

End Results

A few weeks after the launch and following several demonstrations, we analyzed the outcomes: our bot successfully closed 50% of incoming requests.

Now, our employees can use the bot for daily inquiries and find information about product setup or other projects much faster.

We are also scaling our chatbot for client operations, enabling them to find information on how to set up features or the platform without needing to submit additional requests to the client support team.

In our plans, we aim to expand the chatbot solution to address end-user requests and alleviate the support burden for our clients.