The IBM–Illinois Discovery Accelerator Institute (IIDAI) is proud to celebrate a major research achievement: a collaborative team from the University of Illinois Urbana-Champaign (UIUC) and IBM Research has received the Best Paper Award at the 2026 IEEE/IFIP International Conference on Dependable Systems and Networks (DSN) for their paper “PRAXIS: Integrating Program Analysis with Observability for Root-Cause Analysis.”
The award-winning paper was authored by Shengkun Cui (UIUC), Rahul Krishna (IBM Research), Saurabh Jha (IBM Research), and Ravishankar K. Iyer (UIUC) and represents a powerful example of how IIDAI enables impactful academia–industry collaborations that advance both scientific discovery and real-world technology.
Supported by the IIDAI Hybrid Cloud & AI Thrust, the project brought together Illinois researchers and IBM experts to address one of the most challenging problems in modern cloud computing: identifying the root causes of complex software and configuration failures. By combining fundamental research with real-world operational expertise, the collaboration demonstrates the unique value of IIDAI’s model—bringing academia and industry together to tackle ambitious problems that neither could solve as effectively alone.
Modern cloud applications are composed of hundreds or even thousands of interconnected microservices. When failures occur, identifying the root cause can be extraordinarily difficult. Traditional approaches often require engineers to sift through massive amounts of logs, traces, and system data, making diagnosis both time-consuming and error prone. PRAXIS introduces a fundamentally different approach.
The research team observed that cloud incidents rarely occur in isolation. Failures propagate through complex relationships among software components, configurations, and infrastructure. Rather than treating incident diagnosis as a search through unstructured information, PRAXIS explicitly models these relationships using a hierarchical graph that captures:
-
Dependencies among cloud services and microservices
-
Program-level dependencies within individual software components
An AI agent powered by a large language model (LLM) then traverses this graph, following likely fault-propagation paths to identify the underlying cause of an incident. The lead author Shengkun Cui explains:
“One of the key insights of PRAXIS is that agentic root-cause analysis becomes more effective when its exploration is grounded in the inherent dependency structure of the context and task. The hierarchical graph allows context to be assembled dynamically but deterministically, while constraining the LLM to reason along meaningful dependency paths.”
The result is an AI system that reasons more effectively while consuming significantly fewer computational resources. Compared with leading baselines, PRAXIS achieved up to 3.1×-6.3× higher root-cause reasoning accuracy and up to 3.8×-5.3× lower token consumption, depending the LLM model used and the incident scenario.
These results highlight an important lesson for the future of agentic AI systems: success depends not only on more powerful models, but also on providing those models with the right structure for reasoning. By combining advances in program analysis, cloud operations, and AI agents, PRAXIS bridges research communities that have traditionally operated separately. What makes PRAXIS particularly noteworthy is that its impact extends far beyond an academic publication.
The work was showcased at IBM Think, IBM’s flagship business and technology conference attended by thousands of executives, developers, and technology leaders worldwide. More importantly, the core ideas behind PRAXIS are to be incorporated into IBM Concert, where they enhance root-cause analysis capabilities for enterprise cloud environments.
According to co-author Dr. Saurabh Jha, IBM Research scientist and UIUC alumnus:
“PRAXIS has delivered impact well beyond the paper itself. The work demonstrates how AI-driven reasoning can be grounded in software and infrastructure dependencies to produce more accurate and efficient root-cause analysis, while creating a path from research innovation to practical deployment.”
Reflecting on the broader significance of the work, Dr. Alaa Youssef, IBM Co-Lead of the IIDAI Hybrid Cloud & AI Thrust, commented:
“Crucially, the impact is not confined to research only. PRAXIS was showcased at IBM Think, and the underlying ideas are to be implemented as a feature in the IBM Concert product. This is exactly the type of outcome that IIDAI strives to enable—research excellence coupled with real-world impact.”
The project also builds upon a growing portfolio of successful IBM–Illinois collaborations. Previous joint research on autoscaling AI inference workloads has seen adoption across major industry platforms and open-source ecosystems, further demonstrating the long-term value of the partnership.
For Shengkun, who will join IBM soon after he finishes his Ph.D. dissertation, the project was also a transformative research experience.
“PRAXIS exemplifies the core value of the IIDAI program. By bringing together academic research and IBM’s real-world expertise, IIDAI creates an environment where ideas can advance the research frontier while also maturing into practical solutions with impact beyond the paper itself.”
That observation captures the essence of IIDAI’s mission.
Congratulations to Shengkun Cui, Rahul Krishna, Saurabh Jha, Ravishankar K. Iyer, and the entire IBM–Illinois team on this outstanding achievement.