Keeping trains humming along safely and smoothly across Singapore's rail network is a monumental task, especially when engineers have only a three-hour window each night to fix track faults. Now, rail operator SMRT has a new artificial intelligence (AI)-powered tool to help: Jarvis. Playfully dubbed "Just Another Really Intelligent System" by SMRT staff, the intelligent analytics platform was developed by Strides Technologies – SMRT's engineering and tech innovation arm – together with tech giant Oracle.

The Challenge of Night Maintenance

Singapore's MRT system carries millions of passengers daily, making any disruption during operating hours highly disruptive. Maintenance must be conducted in the late-night window when trains are not running, typically between 1 a.m. and 4 a.m. Engineers must use this short timeframe to inspect tracks, signals, power systems, and point machines – the mechanical devices that switch train tracks. Prior to Jarvis, technicians often had to search across hundreds of kilometres of track to locate faulty equipment, a time-consuming process that sometimes leads to incomplete repairs and lingering reliability risks.

The Land Transport Authority of Singapore mandates a strict reliability benchmark known as mean kilometres between failure (MKBF) of one million train-kilometres. Operators must consistently meet this target to ensure minimal commuter disruption. Achieving this requires not only swift fault resolution but also predictive capabilities to prevent failures before they happen. SMRT's existing data – over 30 years of operational, engineering, and failure pattern records – was scattered across multiple systems in text, graphs, and flowcharts, making it difficult to analyse holistically.

Introducing Jarvis: An AI-Powered Solution

Jarvis, announced at the Oracle AI World Tour Singapore on 14 April 2026, addresses this challenge by consolidating decades of dispersed data onto a single platform. It leverages Oracle Cloud Infrastructure (OCI) Enterprise AI and the Oracle Autonomous AI Database to host the data, making it searchable and actionable. The platform uses machine learning algorithms for predictive maintenance and a generative AI (GenAI) chatbot interface powered by large language models (LLMs) and vector search. Engineers can ask natural language questions about specific components, fault patterns, or historical repairs, and receive precise answers in real time.

One of Jarvis's biggest benefits is its ability to convert textual and graphical information into exact geolocation data. As SMRT group CEO Ngien Hoon Ping explained, if a certain fault has been recurring, technicians previously needed to interpret vague reports and search physically along the tracks. Now Jarvis pinpoints the exact point machine or equipment that is acting up. "They go directly to the point machine that same night window and deal with it," Ngien said. "It achieves better effectiveness, high productivity and cost-savings."

How Jarvis Works and Its Impact on Reliability

The platform's predictive models analyse failure patterns to foresee potential breakdowns, allowing maintenance teams to proactively repair or replace components during scheduled night shifts. This reduces the likelihood of in-service failures that delay commuters. The GenAI chatbot serves as a dynamic knowledge base, enabling engineers who may not be familiar with every historical incident to access the collective experience of hundreds of experts. Vector search ensures that even ambiguous queries return relevant data from the stored graphs and flowcharts.

Since Jarvis is in its first phase of deployment, with more than 50 SMRT engineers actively participating, the initial results are promising. Some engineers analyse existing data to refine the models, while others code AI agents to automate repetitive diagnostic tasks. Ngien stressed that the technology is meant to improve the effectiveness of SMRT's workforce of more than 10,000 people, not replace them. "SMRT is still hiring, even in the face of this AI world. We still need engineers," he said. "To us, AI is really about enabling the organisation to uplift our people."

This philosophy aligns with a Kaizen culture of continuous improvement – a Japanese business concept of ongoing, incremental enhancement. Ngien noted that managing a complex locomotive network, from signalling and power systems to railway tracks, requires constant refinement. "It's a very challenging task, even for the engineers among us. But we have this culture to keep improving and make use of the tools available," he added.

Broader Implications for Rail Operations

The challenges SMRT faces are not unique. Urban rail systems worldwide grapple with aging infrastructure, increasing passenger loads, and limited maintenance windows. Predictive AI tools like Jarvis offer a path to more resilient operations. Oracle's senior vice-president for ASEAN and South Asia, Chin Ying Loong, said, "Rail operators depend on timely, accurate data to keep services running safely, reliably and on schedule for millions of commuters each day. Running on OCI, Jarvis demonstrates how Oracle can help bring AI to where enterprise data resides to improve efficiency and operational responsiveness."

Moving forward, Ngien revealed that SMRT hopes to share its experience with other rail operators facing similar challenges. "They also have a trove of data, so through the models we've developed with [Oracle], we would be happy to share with other operators," he said. This open approach could accelerate the adoption of AI-driven predictive maintenance across the industry, reducing costs and improving safety for millions of daily riders.

The deployment of Jarvis also highlights a larger trend: enterprises across sectors are moving AI from experimental proofs-of-concept into production systems that deliver measurable ROI. For CIOs and technology leaders, the SMRT case offers lessons in data consolidation, change management, and the importance of augmenting rather than replacing human expertise. The three-hour night window remains a tight constraint, but with AI, SMRT is making every minute count.

Source: ComputerWeekly.com News