Halyk Agentic SDLC Platform

The initiative was launched to bring AI assistance into the bank’s own technology environment, with the controls required for a regulated financial institution: secure deployment, access management, auditability, data protection, and alignment with internal governance standards.

The main objective was to move beyond individual productivity tools and create a scalable platform for AI agents that can support real delivery workflows across the software development lifecycle. By introducing the platform, the company aimed to automate repetitive and time-consuming work, improve consistency and quality of deliverables, and give teams AI support grounded in real project context from systems such as task management, documentation, code repositories, CI/CD, and testing tools.

In short, the objective was to increase engineering productivity by 30%, reduce manual effort, accelerate delivery, strengthen quality controls, and establish a secure foundation for scaling AI across Halyk Bank.

2026

Quantitatively, the target impact includes:

20-40% productivity improvement in selected engineering and QA activities.

Up to 50% reduction in onboarding time for new team members through project-aware AI assistants.

Around 2x cost efficiency improvement in test preparation compared with fully manual test design.

Reduction of 10+ hours of senior engineer time typically spent supporting onboarding and repeated knowledge-transfer questions.

Potential 90%+ time savings in selected document-heavy or policy-checking workflows once the same platform pattern is expanded beyond software delivery.

For external banking clients, the main benefit is improved speed, quality, and reliability of services delivered by the bank. Faster requirements preparation, better test coverage, and more consistent release processes help reduce defects, accelerate digital product updates, and improve customer experience.

Technology Architecture

The platform is designed as an enterprise-grade AI orchestration layer deployed within Halyk Bank’s controlled infrastructure. It can support cloud or on-premise LLM models depending on data sensitivity and regulatory requirements.

At a high level, the architecture includes:

• AI orchestration layer for routing requests to approved language models, tools, agents, and workflows.

• Model gateway supporting multiple model providers and self-hosted models, enabling the bank to choose the right model for each use case.

• Agent and workflow layer where role-specific assistants, reusable workflows, and automation chains are configured.

• Knowledge and retrieval layer for indexing project documents, repositories, tickets, specifications, and other trusted sources.

• Integration layer connecting to engineering, business, and operational systems through APIs, connectors, and secure tool access.

As foundation we used CodeMe paltform from EPAM Systems which is based on open source components

The project does not use solutions from the Global CIO catalog

Platform implements complex agents Business Analysis and Quality Assurance agents. The BA agent supports generation and refinement of epics, user stories, tasks, bugs and acceptance criteria; validates requirements for completeness and contradictions; identifies duplicate or conflicting backlog items; and standardises ticket structure. The QA agent generates manual test cases, regression and release checklists, and automation inputs from requirements, backend code and API specifications, while supporting test coverage analysis and integration into testing workflows.

The platform is integrated with Jira, Confluence and Allure TestOps, allowing agents to work with real delivery context, project documentation, backlog data and test management processes. Additional use cases can include developer support, code review, DevOps, release management, project onboarding, internal knowledge search, compliance monitoring, HR, legal, IT operations and customer-facing process automation.

The AI Agent Platform initiative introduces an enterprise-grade AI-native delivery platform for Halyk Bank.

The AI Agent Platform provides a governed environment for creating, deploying, and managing AI agents across the software development lifecycle and future business operations. Its core features include:

·        Role-specific AI agents for Business Analysis, Quality Assurance, development, DevOps, code review, release management, onboarding, and internal knowledge support.

·        Business Analysis support: generation and refinement of epics, user stories, tasks, bugs, acceptance criteria, backlog checks, duplicate detection, and requirement validation.

·        Quality Assurance support: generation of manual test cases, automated test scripts, regression checklists, release checklists, and test coverage analysis.

·        Knowledge-based assistance using indexed project sources such as documentation, requirements, code repositories, specifications, and internal knowledge bases.

·        Workflow automation through multi-step AI processes, including handoffs between agents, structured outputs, and human review where needed.

·        Integration with enterprise delivery tools such as Jira, Confluence, Git, CI/CD platforms, test management systems, internal APIs, and document repositories.

·        Governance capabilities including access control, audit logs, usage monitoring, cost controls, model management, and responsible AI guardrails.

·        Marketplace or library of reusable assistant templates that teams can adapt for their own processes.

 Security and governance layer covering authentication, role-based access, auditability, data protection, usage limits, monitoring, and policy enforcement.

·        User interfaces for end users, administrators, and agent builders, including chat-style interaction, no-code configuration, and operational dashboards.

Customisation Capability

The solution is highly customisable. Halyk Bank teams can adapt existing agents or create new ones without full-scale software development. Customisation can include:

·        Defining agent roles, prompts, instructions, tone, output formats, and approval rules.

·        Connecting agents to selected knowledge sources, repositories, business systems, or APIs.

·        Tailoring templates to internal standards for user stories, test cases, checklists, documentation, and release artifacts.

·        Configuring access rights by team, role, department, project, or data sensitivity.

·        Selecting models based on security, quality, latency, cost, or data residency requirements.

·        Platform implements complex agents Business Analysis and Quality Assurance

Kazakhstan