AWS-based generative AI will back first-tier support from Infosys

According to the Glauco Volpe, head of cloud services and infrastructure at Infosys, one of the project’s goals is the smart use of the knowledge produced and stored within the company. “We have two major sources of information for this: the cases already solved by us and application documentation. Combining these two will enable us to solve problems quickly and efficiently,” he explains.

Regarding the use of generative AI, the executive says that pure and simple research tools have lower efficiency and require more experience from analysts. By using generative AI, less experienced professionals can receive work instructions, allowing complex services to be performed at more operational layers.

The development partners were chosen based on their success and partnership track records. “We are global partners of AWS, which is our cloud provider and offers the solutions we provide to our customers. MadeinWeb, on the other hand, has a success story that is deeply connected to the expectations we have with this implementation,” he says.

The solution proposed by MadeinWeb, in collaboration with AWS, was built based on a robust architecture. The process began by creating a secure repository in the Amazon Simple Storage Service (Amazon S3), providing fast, secure, and cost-effective access to thousands of documents related to Infosys calls, knowledge bases, and technical standards.

Native integration with Amazon S3 objects made it easy to implement the Amazon OpenSearch Service, creating a fully vectorized database. This allowed for a more efficient search and advanced analysis of internal calls, ensuring versatility to handle various document formats.

Vinicius Machado, MadeinWeb Cloud, Data & AI director, points out that after extracting this data, the semantic capacity of a foundational model was used to understand the calls and structure the solution to user demands, achieving significant assertiveness in complex problems. Initially, customized models were implemented with Amazon SageMaker, providing substantial computational power. Subsequently, the switchover to Amazon Bedrock brought versatility to the Charla platform in choosing foundational models, streamlining development and experimentation through APIs.

“Choosing Anthropic's Claude-2 foundational model was crucial, due to its advantage in bringing legal concepts into its training, facilitating control over hallucinations and adhering to fairness AI concepts,” explains Machado. Hundreds of calls were analyzed and technical documents processed to ensure the accuracy of Charla's answers. The architecture was then built using key AWS services, including Amazon API Gateway, AWS Lambda Function, and the strategic use of Amazon SageMaker for data ingestion and specific tests.

Volpe explains that the solution will initially be used on only one of the company's clients and then scaled to the entire base. The executive expects the company's first-level service to gain in quality and efficiency by using the solution.

“We now have a tool that will provide solutions to customers and this should result in an increase in volume, a decrease in the number of reopened tickets, an increase in first-call resolutions and a decrease in the number of calls with broken SLAs. These are the main performance indicators that we will use”, she reveals, reinforcing positive impacts such as prompt call response driven by automation and advanced document analysis, culminating in greater operational efficiency.

In addition, the integration between the Charla platform and AWS services reduced the manual burden, allowing the team to focus on more strategic tasks and reducing the SLA on calls. Volpe believes in Infosys’ pioneering role in the adoption of generative AI with AWS. “I would say that we do have somewhat of a pioneering role in treating this solution from the perspective of supporting SAP environments and applications such as Salesforce, Oracle and others for large companies. We are pioneers in directly using AI in service with the creation of a solution designed for production,” he says.