Databricks AI Day Kuala Lumpur 2026 highlights enterprise AI adoption in Malaysia

Databricks AI Day Kuala Lumpur 2026 focused on moving enterprise AI into production with governed data platforms.

Enterprises in Malaysia are moving beyond isolated AI experiments toward governed data platforms that can support analytics, automation, and operational use cases, according to discussions at Databricks AI Day Kuala Lumpur 2026.

The event brought together Databricks executives, PETRONAS, and Malaysia Aviation Group to discuss how organizations are connecting fragmented data systems, applying governance controls, and testing AI agents in production workflows. The sessions covered Databricks products including Lakebase, Genie, Genie Code, Agent Bricks, Lakeflow, and Unity Catalog.

Cecily Ng, Databricks' vice-president and general manager for ASEAN and Greater China, said the company works with more than 20,000 customers globally, including more than 60% of Fortune 500 companies. In Asia, Databricks is working with companies across sectors including utilities, transportation, banking, and telecommunications.

"Across all industries and all sizes, organizations are really leveraging Databricks for faster innovation, better operational decisions, and better customer outcomes," Ng said. She added that Malaysia has become a "vibrant hub" for data and AI activity.

Moving AI into production

A central theme at the event was the difficulty of moving AI from pilots into production. Enterprises often have data spread across transactional systems, analytical platforms, machine learning tools, and streaming systems, making it harder to use information consistently across business workflows.

Kunal Taneja, AVP of Field Engineering at Databricks, said many organizations still struggle despite seeing the potential of AI. "We can see the vision, and we have seen a few people getting a lot of value out of this, but on the whole, organizations are really struggling to make this happen," he said.

Databricks presented its Lakehouse approach as a way to bring analytical and operational data closer together through open formats, shared governance, and common workloads. Delta Lake and Iceberg were cited as open formats for analytical data, while Postgres was discussed as an operational data format for agentic systems.

Lakebase, Databricks' serverless Postgres database, was positioned as part of that operational layer. Agentic systems require a place to store memory, logic, and audit trails, while business users expect low-latency responses from applications.

Governance was another recurring issue. Unity Catalog was described as the governance layer for data, AI models, applications, and access controls across Databricks environments. Taneja said unified governance is not only about securing a database or OLTP system, but about managing workloads and applications more broadly.

Databricks also discussed Genie and Genie Code as part of its AI tools. Genie allows business users to ask questions of enterprise data, while Genie Code supports technical users and data teams working on forecasting, machine learning, debugging, and dashboard development.

Balancing AI ambition with governance

PETRONAS Digital outlined how the national energy company is approaching AI through business value, risk reduction, safety, decision-making, and productivity. Mohd Khairul Zarir Ahmed Lokman, general manager of AI Ecosystem Advancement at PETRONAS Digital, said the company avoids pursuing technology for its own sake.

"At PETRONAS, we don't chase the biggest model, we don't chase the fastest platform, and we don't chase the loudest hype," he said. "We choose a balanced approach that supports business ambition, the needs of our people, and our role as a national energy provider."

PETRONAS is applying AI across upstream, LNG, downstream, trading, and marketing operations. In upstream, AI supports subsurface interpretation and production optimization. In LNG and downstream operations, it is used for plant reliability, efficiency, logistics optimization, and scenario planning.

The company created an AI Ecosystem Advancement team to coordinate AI, data, architecture, governance, security, and trust. Its approach combines internal business expertise with external platforms where the technology ecosystem is already more developed.

Building a shared aviation data layer

Malaysia Aviation Group presented its enterprise data lake, referred to as MABEL, as another example of data platform consolidation. The group uses Databricks to connect data from flight booking, finance, customer surveys, SFTP feeds, APIs, and third-party partner environments.

With Lakeflow Connect, Malaysia Aviation Group copies data from source systems into MABEL. Lakehouse Federation allows it to query data without copying it into Databricks, while Delta Sharing is used to connect with third-party analytical databases.

The group has also tested Lakebase in a proof of concept, with query response times below one second. Data ingestion using Databricks pipelines is 70% faster than its legacy application process, according to the presentation.

Malaysia Aviation Group uses the platform for customer segmentation and customer feedback analysis. Customer segmentation combines passenger behavior, value characteristics, age, purchase patterns, and booking frequency to support targeted marketing across dashboards, websites, mobile apps, and loyalty systems.

Feedback analysis uses structured and unstructured data processed through NLP and AI models inside Databricks. Outputs are fed into dashboards, reports, service adjustments, loyalty systems, and marketing systems.

The group is also exploring Genie for text-to-SQL capabilities, with the aim of enabling staff to retrieve data through natural language queries. Genie Code is being tested to support analysis and dashboarding, while Agent Bricks is being assessed for end-to-end AI workflows.

Using agents for upstream data access

PETRONAS also presented PetroWizard, a multi-agent AI assistant developed for upstream engineers. The tool is designed to address a common problem in upstream operations: engineers often spend more time searching for data than analyzing it.

"Upstream data is everywhere. Answers are not," said Harvard Wong, lead data analyst at PETRONAS. "So we asked, what if an engineer could just ask and get the answer instantly?"

PetroWizard allows engineers to query both unstructured documents and structured operational data. A question about a well's production rate and a related reservoir study can be answered by retrieving data from SQL tables and relevant documents.

The system uses an orchestrator agent to classify user intent and route questions to the appropriate agent. Document-related queries use a RAG agent with hybrid search, combining vector search and BM25 keyword search to handle petroleum-specific abbreviations, well names, field codes, platforms, and equipment terms.

Structured data queries use a Genie-based agent that retrieves information from governed Delta tables. Access control is enforced through Unity Catalog, meaning users receive answers based on their permission levels.

PETRONAS uses LLM-as-a-judge to evaluate PetroWizard responses based on correctness, relevance, and whether the right tool was used. The company completed a proof of concept and user acceptance testing last year with 20 upstream participants, achieving an average score of 4.25. PetroWizard is now being piloted in the upstream domain, with additional use cases being assessed around project risk lessons learned and mitigation based on past development projects.