Building an AI-powered customer rankings platform on AWS with Centras Group

Centras initially set out to analyze customer feedback for one of its restaurant chains. To understand how customers reacted online to competing brands across maps and social channels, Centras had built an in-house machine learning model that’s based on RoBERTa—a high-performance transformer model—to translate reviews, detect sentiment, and classify feedback into predefined categories.

While that model produced useful insights for a single business segment, scaling it across additional industries proved challenging. Each new segment required retraining the model and manually tagging between 30,000 and 50,000 reviews. Supporting multiple industries simultaneously became operationally unsustainable, with significant internal resources dedicated to model maintenance rather than insight generation.

To move beyond these constraints, Centras evaluated managed cloud services and turned to AWS. Generative AI offered a way to analyze feedback across industries without rebuilding or retraining the solution for each new use case.

Centras adopted Amazon Bedrock for building generative AI applications and agents at production scale. The company selected Amazon Nova foundation models—which offer frontier intelligence and industry-leading price performance—to process multilingual customer feedback with high accuracy and flexibility.

Centras collects and uploads public customer reviews from maps, review sites, and social channels to Amazon Simple Storage Service (Amazon S3), object storage built to store and retrieve any amount of data from anywhere. Amazon SageMaker AI—which is used to build, train, and deploy AI models—performs data cleaning and filtering and prepares the data for further analysis.

The prepared data is then processed on Amazon Bedrock using the Amazon Nova Pro foundation model. The model handles translation between Kazakh and Russian, performs sentiment analysis, applies automated tagging, and categorizes feedback across more than 50 standardized themes. Processed results are stored in Amazon S3. Amazon SageMaker then calculates performance metrics and prepares analytical outputs that feed directly into Centras Rankings and the C500 framework.

By redesigning its architecture around managed services on AWS, Centras shifted its focus from maintaining machine learning infrastructure to generating scalable customer experience insights across industries. Consolidating translation, sentiment analysis, and categorization into a managed generative AI layer reduced reliance on manual tagging and minimized the need to retrain multiple custom models.

“We realized that Amazon Nova Pro could replace a large part of our internal machine learning workload, including translation, sentiment analysis, tagging, and issue categorization,” says Daniyar Kazhenbayev, managing director at Centras.

More than the technology itself, Centras valued the role AWS played in helping the team focus on the business problem rather than the underlying complexity. “The AWS team wasn’t focused on selling technology but on helping us succeed,” says Kazhenbayev. “We work with many global tech providers, but AWS was one of the first to try to understand our motivations and goals.”

The new approach delivered clear, measurable results. Centras reduced manual review and data tagging work by up to 90 percent, freeing internal teams to focus on market analysis and industry benchmarking rather than operational processing. Where the team previously analyzed one business segment in about 2 days, it can now analyze more than 10 segments in the same time frame.

The solution consistently identifies more than 50 standardized issue categories and supports analysis in both Kazakh and Russian. It’s currently used internally to analyze industries including quick-service restaurants, banking apps, insurance, healthcare, and professional services.

Beyond efficiency gains, Centras’s solution has created new strategic value. By combining customer sentiment insights with the C500 financial rankings, Centras created a unique analytical framework linking brand perception with financial performance across industries—a capability previously unavailable in the Kazakhstan market.

Looking ahead, Centras plans to move the solution fully into production and offer analytics services to external organizations through dashboards and predefined industry reports. “This project is not just a tool for individual companies but also a platform for positioning Centras as a standard-setter for customer experience analytics in Kazakhstan,” says Kazhenbayev.