Reliable AI Systems
Methods for measuring capability boundaries, detecting regressions, and limiting the impact of failures.
Research / Systems that can be trusted
Reliability across software, data, and AI.
My interests sit at the intersection of system reliability, data-intensive systems, and AI engineering.
01
Current interests
LLM Evaluation
Evaluation design that connects offline test sets, production traces, and real task outcomes.
Agent Reliability
Failure analysis for long-running workflows composed of planning, tools, memory, and external state.
Streaming Data Systems
Event-time reasoning, provenance, runtime monitoring, and recovery in stateful stream processing.
02
Previous research themes
- Test-data generation for stream-processing systems
- Fine-grained data provenance
- Runtime monitoring and failure prediction
- Reliability modeling for data-intensive software
03
Research output
Publications, talks, and reproducible research artifacts will be added here with source links. This page intentionally avoids unverified placeholders.