Codebit faced a major challenge with the quality of communication between its employees and clients. Centralized in its internal platform SGM - Maintenance Management System, the exchange of messages was frequently hindered by grammatical errors, lack of clarity, conciseness, and objectivity.

In search of assertiveness, Codebit turned to AWS Generative Artificial Intelligence to transform this scenario. Through a carefully architected solution, the company integrated advanced linguistic correction and automatic summarization features, promoting a more professional, efficient, and comprehensive communication.

About the client:

Codebit Desenvolvimento de Softwares Customizados LTDA, a Brazilian company founded in 2010, operates in the technology sector and specializes in customized software development and cloud infrastructure management, focusing on AWS - Amazon Web Services services. 

With a team of more than 70 employees, the company manages its internal operations through its own platform called SGM - Maintenance Management System. 

Operating in the medium-sized business segment, Codebit serves highly relevant clients, such as Samsung, Fundação Itaú, Doctors Without Borders, and government bodies linked to the National Council of Justice.

Client Challenge:

Communication with clients is entirely centralized in the SGM platform, where employees interact, request information, access history, and monitor maintenance demands. Each request, or "ticket," represents a demand, often filled with comments that require constant alignment. 

However, this process presented many challenges, such as grammatical errors, poor linguistic structuring, and inappropriate tone in messages, which negatively impacted clients' perception of the company's technical competence. In addition, reading long conversations hindered understanding and operational efficiency.

Given this, Codebit identified an opportunity for innovation through the use of Generative Artificial Intelligence. The proposed solution included two main features: a comment correction engine with tone adjustment, and an automatic summarization tool. The first aimed to ensure that messages were correct and conveyed professionalism, while the second synthesized the main points of each demand, facilitating comprehension of the history.

Partner Solution:

This initiative was framed within the AWS Generative AI Consulting Services specialization. The solution was custom-developed, integrating into the SGM platform and improving the user and client experience. The adopted architecture prioritized security, scalability, and resilience. 

With resources distributed between public and private subnets, the environment was orchestrated via AWS Fargate containers, using Anthropic's Claude Sonnet V3.7 model through Amazon Bedrock for correction and summarization tasks. Version control and automated deployment were implemented with AWS CodePipeline, CodeBuild, and CodeCommit/GitHub.

The system was designed to ensure fault tolerance, with automatic restart of tasks in case of error and the use of Bastion Hosts for security in accessing private resources. Although a multi-AZ architecture was not adopted due to cost considerations, the structure remains prepared for future expansions.
Integration with the client's platform occurred through two main paths: a comment correction API, triggered before the message was registered in the system; and an embedded chat module by JavaScript injection, capable of interacting in real-time with open demands. Both interfaces use custom prompt engineering, and records are stored in Amazon RDS for monitoring and continuous improvement.

Several alternative language models were tested before defining Claude Sonnet as the standard. Models like LLaMA and Claude Haiku were discarded because they did not meet the requirements for precision and tone refinement. Version 3.5 of Sonnet showed the best results, and was subsequently updated to version 3.7, ensuring even more fluidity and contextual fidelity.

Results and Benefits:

In production, the solution has shown a significant impact on Codebit's operations. Within one month, thousands of comments were processed and hundreds of users benefited from optimized summaries and responses. From a technical standpoint, the system processes hundreds of requests and millions of tokens daily, contributing directly to AWS annual recurring revenue (ARR).

From a strategic standpoint, the adoption of Generative AI brought measurable gains, such as a 35% reduction in time spent reading comments and a 20% decrease in rework and communication failures. In addition, internal surveys indicated that 90% of employees are satisfied with the tool, highlighting time savings and improved communication clarity.

However, an identified challenge was latency in comment processing. Although small delays caused some frustration, it was decided to prioritize response quality over speed, keeping the Sonnet 3.7 model as the default. 

The application of artificial intelligence to transform the company's internal and external communication represents a milestone of innovation and efficiency, with a direct impact on service quality, client perception, and team productivity.

With measurable gains, such as reduced message reading time and fewer communication failures, the quality of interactions with clients has significantly increased.

The solution, based exclusively on native AWS services, not only reinforces Codebit's technical competence but is also aligned with AWS's Generative AI competency. For the future, Codebit plans to expand the solution with monthly management reports, offering an executive view of operations.