What can intelligent systems prove about their own work?
Research in self-verifying agents, AI-assisted query optimization, and autonomous cloud data infrastructure.
Current work spans self-verifying agentic systems and AI-assisted optimization across queries and cloud infrastructure.
AI-assisted query rewriting, program synthesis, and autonomous cloud-warehouse optimization.
Uses large language models and counterexample-guided validation to rewrite SQL queries while preserving semantics.
Learns workload-aware policies for cloud-warehouse performance, resource sizing, and cost optimization.
AI systems that can test, diagnose, and demonstrate when their work is actually complete—and preserve consistent assumptions across durable workflows.
Making database behavior, latency, resource use, and uncertainty more predictable.
Databases that get smarter with every workload.
From bounded approximation to portable, interactive analytics over massive data.
Research on workload-aware cloud data-warehouse optimization was commercialized through Keebo, whose published system applies learning to performance, resource sizing, and cost decisions with operational safeguards.
SnappyData commercialized ideas introduced by BlinkDB in a unified platform for streaming, transactions, and interactive analytics. TIBCO acquired SnappyData in 2019 and introduced the technology as TIBCO ComputeDB.
Research from Mozafari’s group produced VATS and CATS, contention-aware scheduling techniques adopted by MariaDB and MySQL. MySQL/InnoDB currently uses CATS; MariaDB made VATS its default scheduler in version 10.2.3 and later removed it.
DBSeer made workload analysis, resource/performance prediction, and database diagnosis available as public software for MySQL and PostgreSQL.