Work
Production AI systems, upstream work in the open-source infrastructure they depend on, and the research that connects them.
Selected work
- AI platformsProductionA consumer-facing AI assistant on Amazon Bedrock AgentCoreAgentic assistant with grounded retrieval, governed tool access, and conversation memory, plus the shared AI platform internal tools now build on.
- Platform engineeringProductionCloud-native microservices for mortgage loan automationDesigned and built many of the platform's scalable microservices on AWS, owning each from design and Terraform to code and deployment.
- AI infrastructureOpen-source projectkvfleet: cache-aware routing for LLM fleetsA routing control plane that keeps KV-cache locality, policy, and health in one explainable decision.
- Data systemsUpstreamCheckpoint safety for change-data-capture in Apache SeaTunnelPinned down what a CDC checkpoint is allowed to contain when a fetcher fails mid-handoff.
- Query enginesUpstreamFaster, bounded query planning in Apache DataFusionRemoved two planner hot spots: repeated name resolution on wide queries and runaway regex compilation.
- Build platformsUpstreamNew public extension APIs for Apache Maven 4Gave Maven extensions stable, Maven-owned contracts for repository events and settings parsing.
- ResearchEventFlowSentry: reproducible fault testing for streamingTurns event-time failures in streaming pipelines into reproducible, explainable experiments.
Open-source tools I maintain
Small, focused libraries for AI systems, testing, and developer workflows.
- bigocheckZero-dependency empirical Big-O complexity checker for Python with CLI, assertions, and pytest integration.
- kvfleetKV-cache-aware routing control plane for self-hosted and hybrid LLM fleets.
- mcp-egress-guardPolicy enforcement proxy for MCP tool calls with leak detection, blocking, redaction, and audit logging.
- mcp-poolAsync connection pool for Model Context Protocol client sessions.
- mcp-shield-piiReal-time PII redaction proxy for Model Context Protocol workflows.
- pdperfStatic performance linter that detects slow Pandas anti-patterns before production.
- pydanticforgeInfer robust Pydantic v2 models from messy, evolving JSON streams.
- schemaglowHuman-friendly schema diff and contract drift detection across data files and API specs.
- toolcallcheckDeterministic Python testing for tool-using agents and MCP workflows.
- tracemapTraceroute visualization for terminal and web with ASN, GeoIP, replay, diff, and privacy-aware output.
- vectormigrateSafe embedding and vector-index migration tooling for retrieval systems.
- promptspecjJava prompt specification toolkit for Spring AI and enterprise LLM workflows.
Also contributed to
Apache Spark, Apache Beam, Apache Pinot, Apache Arrow, CPython, Go, Rust Cargo, PyTorch ExecuTorch, ClickHouse, Spring AI, Spring Cloud Gateway, SQLAlchemy, Meta Pyrefly, OWASP, OpenTofu, Kubernetes inference-perf.
Full history on GitHub