English | 한국어

LLM AgentOps Engineer

I specialize in LLM AgentOps—turning agentic prototypes into observable, testable, cost-aware, and reliable production systems. I also build manufacturing planning platforms where complex operational workflows must become validated and explainable decisions.

I design Claude and Codex agent harnesses that turn open-ended development tasks into controlled, repeatable, and reviewable engineering workflows.

• Define scoped context and task boundaries so each agent receives the minimum information and authority required for its role.
• Configure tool permissions and controlled execution paths for repository, browser, document, and delivery workflows.
• Orchestrate plan → implementation → test → review → PR workflows with explicit checkpoints and recovery paths.
• Use verification gates, reviewable artifacts, and traceable handoffs to make outcomes reproducible and auditable.
• Separate implementation and review roles when independent validation improves confidence.

A production AI-agent platform where LLM AgentOps practices make natural-language analytical workflows observable, validated, cost-aware, and reliable for advanced planning and scheduling.

• Built and refined multi-step agent behavior across planning, SQL-assisted analysis, validation, replanning, and report generation.
• Centralized LLM execution paths so runtime context, token usage, inference traces, and cost signals remain consistent across agent workflows.
• Added regression coverage for prompt injection, execution context, and SQL tooling to protect behavior during refactoring.
• Expanded production observability with Prometheus metrics and Grafana dashboards across agent, API, database, model, session, and tool layers.
• Contributed operational controls for usage reporting and protected API documentation access.

A manufacturing planning engine for order allocation, machine scheduling, and lot composition in corrugated-packaging operations.

• Prevented stale demand from being processed twice during lot composition.
• Fixed confirmed-order recomposition flows so eligible orders remain in factory-specific planning.
• Refactored coating domain rules and strengthened work-in-progress and order-length data integrity.
• Added repeatable QA coverage with unit tests, integration datasets, and GitHub Actions workflows.
• Improved operational support with Oracle schema/index versioning, configurable logging, and engine-result exports.

Software Developer at BeaconFire Inc., contributing to a real-time fund-processing platform for financial services.

• Strengthened availability in OpenShift/Kubernetes by resolving liveness-probe issues, reducing recurring health-check failures from 40 per hour to zero.
• Designed multi-process, event-driven services with Kafka and AMQ/RabbitMQ across Azure SQL, PostgreSQL, and DynamoDB, and refactored legacy code into an order management system.
• Optimized database-intensive batch processing for 100,000+ orders through caching, batching, and indexing.
• Supported Azure disaster recovery and failover, Sumo Logic/Log4j2 monitoring, and Veracode vulnerability remediation.
• Established Docker/Flyway test environments and applied TDD with JUnit 5 and Mockito to improve deployment stability.

Full-Stack Engineer at VisionSpace, building a hybrid-cloud digital-twin service platform for warehouse operations.

• Designed Java/Spring Boot microservices across on-premise and AWS infrastructure, with Jenkins delivery pipelines targeting EC2, ECR, and ECS.
• Built a React/TypeScript frontend and web application server for visualizing and interacting with dynamic digital-twin simulations.
• Developed a .NET/C# stress-testing application that simulated high-concurrency robotic-device traffic through MQTT, WebSocket, and AWS IoT.
• Integrated LLM, RAG, and computer-vision capabilities with the platform to expand intelligent search, analysis, and visual processing workflows.
• Supported production connectivity and scaling with Route 53, ACM, S3, security controls, and load-balanced AWS services.

• ModelTranslator — model-based IoT tooling that translates UPPAAL models into Python-oriented embedded-system workflows.
• Hibernate quiz web app — a Spring Boot and Hibernate web application for working with relational data and protected routes.
• D* algorithm — pathfinding exploration implemented with Python/Pygame and Java desktop UI tooling.