As artificial intelligence reshapes the foundations of enterprise software, financial institutions are rethinking how data platforms can support more intelligent, automated and business-oriented operations. In an executive interview with the China Academy of Information and Communications Technology (CAICT) under its Zhuji digital transformation initiative, Yulin Cai, President of Data Business at Sunline, outlines how DataOps is transitioning from static data infrastructure into an autonomous value pipeline.
The conversation examines how DataMind, Sunline’s self-developed integrated data intelligence platform, combines DataOps, lakehouse architecture and AI to help financial institutions move from AI experimentation towards production-ready applications and measurable business outcomes.
Q1: As AI changes the way financial institutions use data, how is DataOps evolving from traditional infrastructure into an intelligent value pipeline?
Yulin Cai: This transformation operates across two complementary dimensions:
• DataOps as the AI Foundation: DataOps provides the automation methodology required to elevate raw lakehouse storage into trusted, governed data assets. By integrating multi-modal storage, real-time consumption, and enterprise semantic modeling, institutions inject domain business context directly into data pipelines. This contextual foundation bounds AI reasoning, curbing model hallucinations and ensuring regulatory compliance.
• AI-Driven DataOps Automation: Embedded AI capabilities automate schema design, code generation, testing, maintenance, and data governance. This elevates DataOps from a manual, labor-intensive operational workflow into an agile, self-evolving value pipeline.
DataOps and AI are mutually reinforcing: DataOps delivers the governed data foundation AI requires, while AI injects automated intelligence back into DataOps - redefining the value boundaries of modern enterprise data platforms.
Q2: DataMind natively integrates Agent modules into its core platform capabilities. What is the core driving logic behind this architecture, and how is it structured?
Yulin Cai: The starting point is a practical one: AI needs to be connected to reliable data, domain knowledge and enterprise workflows before it can create sustainable business value.
DataMind therefore does not treat AI agents as a separate add-on. Agent capabilities are integrated into the platform alongside data engineering, data governance and analytics.
The architecture combines Base LLM with Domain Knowledge Injection, allowing organizations to build specialized agents for specific financial workflows. Each agent operates within a dedicated sandbox environment, where it can invoke tools, execute tasks and perform self-correction under defined controls.
This is particularly important in financial services. An enterprise agent cannot simply generate an answer; it needs to operate within appropriate business rules, data boundaries and governance requirements. The combination of DataOps, lakehouse and AI provides the technical foundation for making AI more trustworthy, traceable and operationally controllable.
Ultimately, the objective is to move AI from isolated experimentation towards integration with real enterprise processes.
Q3: Enterprise AI deployment faces widespread bottlenecks in model hallucinations, task planning accuracy, and single-agent execution limits. How does Sunline overcome these technical hurdles?
Yulin Cai: One of the key lessons is that reliable enterprise AI is an engineering problem as much as a model problem.
Single agents fail when attempting to bridge complex, end-to-end banking processes - such as customer segmentation, strategy design, campaign execution, and performance optimization. To overcome this, DataMind utilizes a Multi-Agent Orchestration Architecture to divide labor among specialized agents collaborating across complete workflows.
To maintain deterministic reasoning, Sunline embeds two architectural mechanisms refined across 400+ financial institution implementations:
• Domain Knowledge Guardrails: Encapsulated data models, validation rules, and metric definitions establish a strict business constraint layer that bounds model reasoning and mitigates hallucinations at the source.
• Standardized Skill Framework: Expert operational workflows are packaged into reusable execution modules ("Skills"), reducing foundational model variability while improving task planning and execution accuracy.
Q4: Where do you see AI adoption in financial services today? What measurable value can institutions achieve from real-world implementations?
Yulin Cai: AI has shifted from a passive productivity assistant into an active participant in business operations, where industry focus has moved from foundational model capabilities to deployment efficiency and concrete operational ROI.
Our implementation experience shows that this can translate into measurable operational improvements:
• In data engineering, DataMind uses a multi-agent architecture comprising an orchestration agent, specialist agents covering offline development, real-time development, batch scheduling, data services and intelligent operations, together with an audit agent. This creates an end-to-end workflow from natural-language requirements to automated deliverables and has improved overall data development efficiency by 70%.
• In data governance, AI is applied across project preparation, rule generation and execution, and reporting. It can identify data quality issues, analyse root causes and recommend remediation actions. In one implementation, rule discovery efficiency improved by 66.7%, while the accuracy of remediation recommendations reached 80%.
• For business intelligence and data insights, agents can perform tasks such as analysing indicator fluctuations and conducting DuPont analysis, then generate visual reports for management. This has improved indicator interpretation and analysis efficiency by 90%.
Q5: Expectations for enterprise AI are moving from "efficiency tools" to "decision collaborators." What strategic milestones must be crossed to transition DataMind from a software platform into an intelligent business partner?
Yulin Cai: Evolving from a transactional software tool into an enterprise decision partner requires crossing three strategic milestones:
First, establish a unified semantic command layer between data and AI.
Financial institutions often have data, metadata, business standards and security policies distributed across different systems. Bringing these elements together into a unified semantic layer can give AI a clearer understanding of enterprise data and business context.
The longer-term goal is to make DataOps accessible through natural language, allowing business users to express what they need in everyday language while the platform translates those requirements into governed data operations. In other words, natural language becomes a command interface for enterprise data.
Second, rethink the model for technology delivery.
Traditional financial technology projects rely heavily on human resources across solution design, architecture, project management and implementation. AI creates an opportunity to redesign this model by introducing AI agents as part of delivery teams, working alongside consultants and engineers while maintaining appropriate technical and business governance.
Third, industrialize the development and deployment of AI agents.
Instead of building every agent as a project-specific solution, financial institutions need a more standardized Agent Factory approach. Orchestrated agents can take responsibility for activities such as software development, testing, data migration and quality validation, while human teams focus on architecture, business decisions and higher-value innovation.
Therefore, DataMind not only aims to become a more intelligent data tool but to evolve into an intelligent business partner - one that brings together trusted data, governed AI and financial domain expertise to help institutions make better use of their data and turn AI capabilities into sustainable business value.
About CAICT
The China Academy of Information and Communications Technology (CAICT) is a leading research and advisory institution under China’s Ministry of Industry and Information Technology (MIIT), focusing on information and communications technology, digital transformation and emerging technologies. It plays an important role in technology research, industry development, standards and policy research across China’s digital economy.
