Systems
Systems for evolving agent working environments, semantic queries, graph databases, and accelerator-native execution.
My systems research spans external working environments for AI agents, semantic query processing, graph databases, and accelerator-native execution. These are distinct systems and research lines, not components of one deployed stack.
Agents & retrieval
A self-evolving harness in which agent teams organize data from files, databases, and apps, construct reusable context, memory, tools, and skills, and evolve the external working environment through task feedback.
A production AI search engine for trillion-scale vector data, combining graph-index construction with hybrid memory and SSD storage.
Query & graph engines
A semantic query engine built on DuckDB, integrating LLM-powered operators into SQL with query optimization and adaptive execution.
A production graph database service combining composable architecture and factorized execution for high-concurrency interactive graph workloads.
A tensor-based SQL query engine using tensor runtimes as a portable execution layer across heterogeneous accelerators.
A tensor-centric graph processing framework with graph operators, compression, and out-of-memory execution across tensor runtimes and accelerator backends.
A PyTorch-based graph query engine using compact tensor representations of graph topology and batched graph-query operations.
A vector-based engine for scalable GPU-accelerated subgraph query processing.
Distributed analytics
A distributed SQL engine that separates communication from local computation to expose a larger query-optimization space.
A distributed graph analytics system that expresses subgraph counting as relational queries, reusing query optimization and execution techniques.
A compressed subgraph-matching system that computes compressed results directly to reduce intermediate and output materialization.