Agent execution
Multi-tool planning and replanning, dependency resolution, validation, evaluation, safeguards, and human approval.
AI agent platforms · Distributed systems
I build the foundations that let AI agents act reliably in the real world.
I translate customer needs into reusable systems for building, orchestrating, evaluating, and safely operating AI agents in production. My work spans agent building, agent harness, integrations and tools, MCP, permissioning, governed execution, and reliability.
A systems builder—from optimizing distributed systems to architecting resilient AI platforms.
Current work
Production agents need more than intelligence.
They need governed access to tools, durable execution, clear permission boundaries, and reliable recovery when external systems behave unpredictably.
Capability creation
No-code creation of validated, test-run tools from API specifications, documentation, MCP servers, and other system descriptions.
Enterprise action layer
Reusable integrations, authentication, permissions, credential isolation, governed execution, and auditability across external systems.
Workflow reliability
Schedules, webhooks, polling, durable state, deduplication, re-drive, partial-failure recovery, and production observability.
How I work
Architecture that becomes a reusable product capability.
Start with the real workflow. Understand what customers and operators need to accomplish across existing systems.
Design the shared foundation. Turn one requirement into an abstraction that works across products and tenants.
Drive through production. Set technical direction, mentor engineers, review architecture, and own operational outcomes.
Research
Resilient systems for critical infrastructure.
My doctoral work develops real-time distributed systems that preserve correct operation through component compromise, network attacks, and compound cyber-physical threats.
Ph.D. advisor: Dr. Yair Amir
Read my Ph.D. thesis: Real-Time Byzantine-Resilient Power Grid Infrastructure ↗01 · Real-time distributed systems
End-to-end Byzantine-resilient power-grid infrastructure
An integrated resilience framework for broad grid SCADA operations—not isolated protection of a single component or subsystem.
Research gap
Prior intrusion-tolerant SCADA research largely addressed isolated concerns, such as control-center replication or a single component failure. It lacked a comprehensive threat model and an integrated solution spanning the broader grid infrastructure and its varied real-time operations.
Architecture and contribution
My thesis introduced the first end-to-end Byzantine-resilient system framework spanning control centers, substations, RTUs, PLCs, and field devices. The architecture combines replicated SCADA masters, Byzantine-resilient substations, protected field devices, and intrusion-tolerant networking under simultaneous system compromises and network attacks.
Operations and validation
I demonstrated control-center commands, substation commands, autonomous substation protection, and status updates in an integrated real-time scenario. Separately, the real-time substation architecture was evaluated across extensive operating conditions and in a Sandia National Laboratories purple-team exercise at PNNL using protective relays from Siemens, GE, and Hitachi Energy; the technology was transitioned to GE, Siemens, PNNL, and Sandia.
02 · Cyber-physical resilience
Compound-threat analysis and tolerance
Understanding what happens when severe physical disruption is followed by a coordinated cyberattack against already-degraded critical infrastructure.
Systems challenge
Natural disasters can damage infrastructure and communications while forcing emergency reconfiguration. A capable attacker can exploit that degraded state, turning individually survivable events into a compound threat with cascading consequences.
Research contribution
I helped define the compound-threat model and developed methods to analyze how cyberattacks interact with physical disruption across control systems, networks, and recovery mechanisms. The work exposed weaknesses in static architectures and motivated more adaptable resilient-system designs.
From analysis to resilience
The research connected data-centric impact analysis with system-level protection and recovery, informing reconfiguration capabilities in Spire and a broader approach to severe-impact-resilient control systems. It received the IEEE SRDS 2024 Best Paper Award.
Open source
Systems built to keep operating under attack.
Research and production-oriented infrastructure for intrusion tolerance, Byzantine replication, and resilient networking.
Selected publications
Distributed systems, cybersecurity, and critical infrastructure.
- 2024 · IEEE SRDS · Best PaperTolerating Compound Threats in Critical Infrastructure Control Systems
- 2024 · IEEE PESGMByzSec—A Multi-layered Byzantine Resilient Architecture for Bulk Power System Protective Relays
- 2022 · IEEE RTSS WorkshopReal-Time Byzantine Resilient Power Grid Infrastructure: Evaluation and Trade-offs
- 2022 · IEEE SRDSReal-Time Byzantine Resilience for Power Grid Substations
- 2021 · IEEE/IFIP DSN WorkshopData-Centric Analysis of Compound Threats to Critical Infrastructure Control Systems
Experience
From performance engineering to resilient AI systems.
Ema · Software Engineer
AI agent-platform architecture, integrations and external actions, agent harness, workflows, MCP, governance, evaluation, and production reliability.
Johns Hopkins University · Doctoral Researcher
Real-time Byzantine-resilient systems, critical-infrastructure protection, cybersecurity, and ML-based intrusion detection.
Aktrix Technologies · Software Engineer
Performance engineering leadership, automated monitoring, operational data pipelines, and predictive analysis.
Deloitte · Performance Engineering Analyst
Performance testing and optimization of geo-distributed enterprise systems.
Teaching
Seven course appointments.
Machine Learning, Advanced Distributed Systems, Distributed Systems, Software for Resilient Communities, Intermediate Programming, and Introduction to AI at Johns Hopkins and Duke.
Contact