Bank CenterCredit builds a secure hybrid architecture for AI using AWS

BCC is a fast‑growing and actively developing bank, and its demand for computing resources and new technologies was increasing rapidly. The bank understood that its on-premises infrastructure was a limitation—both in terms of available computational capacity and the speed at which BCC could scale and modernize. Obtaining new servers would take months, and maintaining the capacity required to support peak demand would result in high operational costs. To meet its growing resource demands, the bank looked to cloud technologies.

Global cloud providers offer a wide range of unique, ready‑to‑use services across different domains. BCC knew that adopting these services would be significantly faster and more efficient than developing similar solutions entirely in house. “We opted to build a hybrid, multicloud architecture. This approach provided us with flexibility in scaling computational resources as well as the ability to use unique cloud services to address various technological challenges and use cases,” says Darkhan Aspandiyarov, vice president of IT at BCC. BCC selected AWS Outposts so that it could establish a hybrid architecture while using the bank’s own data center, which would help maintain compliance with regulatory requirements such as data encryption and protection.

The architecture centers around using AWS Outposts to run AWS services locally. Using this solution, BCC can use a subset of AWS services locally to preprocess regulated data that must remain on premises while accessing a broader set of AWS services in the parent AWS Region to develop generative AI solutions. These services include Amazon SageMaker AI to build, train, and deploy AI models and Amazon Bedrock to build generative AI applications and agents. For encryption, BCC uses AWS Key Management Service (AWS KMS) External Key Store(XKS), making it possible for BCC to store encryption keys in its on-premises data center. BCC manages cryptographic keys in virtual security modules that are external to the AWS environment and deploys AWS KMS XKS proxies on AWS Outposts, which mediates all communication between AWS KMS in the parent AWS Region and the on-premises security modules. This means that AWS KMS does not interact directly with the security modules—providing BCC sovereign control over its regulated data and meeting regulatory requirements.

“Aligning with regulators and auditors on encryption, access controls, and secure hybrid development took time, but using AWS Outposts, we met these requirements without compromising the speed of innovation,” says Rubina Lozovaya, vice president of data management at BCC.

Using its new hybrid cloud architecture, BCC developed several generative AI solutions that benefit both customers and BCC’s teams, including an ASR solution to manage customer service calls in both Russian and Kazakh. To train the foundation model for multilanguage ASR, BCC loaded the audio files of over 10,000 historical customer service calls to Amazon Simple Storage Service (Amazon S3)—an object storage service—on AWS Outposts, preprocessing the call data locally to remove personally identifiable information such as customer name and address. BCC processed the calls using Amazon Elastic Compute Cloud (Amazon EC2), which provides secure and resizable compute capacity, then cleansed and prepared the data that was seamlessly replicated to Amazon S3 in the parent AWS Region, using Amazon SageMaker AI to improve domain-specific speech recognition. (See figures 1, 2, and 3 below.)

With this enhanced ASR model, BCC was able to increase the transcription accuracy of audio data from 81 percent to 90 percent for customer service calls in Russian, from 74 percent to 89 percent for customer service calls in the local Kazakh language, and from 63 percent to 86 percent for customer service calls in mixed Russian–Kazakh.

Using this architecture, BCC also developed an internal HR chatbot. Core to the HR chatbot was the implementation of local retrieval augmented generation (RAG), which helps BCC generate responses based on employee information that must remain on premises. BCC deployed embedding models on premises to transform user queries and on-premises HR knowledge bases into vector embeddings that are stored in a vector database running on AWS Outposts. The vector database performs a similarity search to generate responses in Russian or Kazakh, depending on user input. This solution significantly improved HR efficiency while helping BCC meet regulatory compliance. The solution also created a blueprint for automating internal processes in IT support, security, documents, and finance.

With a solid digital foundation built on AWS, BCC was able to unlock even greater possibilities by using AWS AI services in the parent AWS Regions. For example, BCC also developed a solution using Amazon Bedrock to summarize and filter financial news through batch inference. Now, the bank automatically summarizes financial headlines and extracts key insights to help analysts react faster to market changes while reducing manual workload.

“We achieved the best of both worlds—full compliance and data residency within Kazakhstan, combined with the scalability and innovation of the AWS Cloud,” says Lozovaya. “This hybrid architecture empowers us to run advanced AI and machine learning workloads securely and deliver new digital services faster.”

Using AWS to support its generative AI solution development, BCC is accelerating innovation while optimizing costs. The bank’s ASR solution processes more than 1 million minutes of audio monthly with a cost savings of over 70 percent compared to its previous ASR provider. The solution has helped BCC reduce its word error rate by 55.8 percent and increased recognition accuracy by 22.6 percent. And by using Amazon Bedrock in its news-processing solution, BCC reduced AI inference costs by 50 percent.

BCC will continue to develop generative AI solutions using its secure, compliant architecture. “Using AWS Outposts facilitated compliance with the strict security and data residency requirements specific to the banking sector in our country,” says Aspandiyarov. “Ultimately, this approach provided a balance between innovation, performance, and regulatory compliance, which is critically important for a financial organization.”