Interactive resume system
Keith Swoger
AI agent engineering leader building production-minded agent systems: workflow translation, retrieval (RAG), tool use, memory, governance, evaluation, and human-in-the-loop operations.
Current project: AgentReef + Angel Software direction
AgentReef Operations
AgentReef is the project where I am shaping agent operations into a real control plane: intent routing, planning, retrieval, tool access, memory, governance, evaluation, and human approvals in one accountable surface.
- Project
- AgentReef and Angel OS systems for coordinator, security, evaluation, memory, retrieval, and human-approval surfaces.
- Technical focus
- Intent routing, scoped tool access, RAG patterns, durable context, model selection, prompt and version governance, and traceable execution.
- Evidence
- Years of operations work where reliability, stakeholder visibility, escalation, and accountable delivery matter more than demos.
system agents checked in
dispatch policy route selected
human approval gate armed
trace evidence linked
Trust requires competition and measurement
Evaluation Arena
This section is based on Arena, my evaluation project for testing agent behavior. The goal is simple: agents should be tested against scenarios, regressions, latency, cost, and safety expectations before they earn more autonomy.
- Experience
- Release validation discipline: scenario definition, manual and automated test execution, bug reporting, KPI awareness, and regression control.
- Agent lesson
- Agents should earn autonomy through evals, benchmark cases, failure analysis, accuracy/safety measurement, and recovery behavior.
- Evidence
- Apple QA release validation across 10+ OTA updates, plus high-volume diagnostics and customer issue resolution.
Aquarium inspired observability
Behavior Aquarium
This section maps to Aquarium, the observability project for understanding agent behavior. It turns data access, memory, tool calls, handoffs, and blocked decisions into something an operator can inspect.
- Purpose
- Agent operators need to see why work moved, what data was used, which controls applied, and when a human decision became necessary.
- Technical focus
- Behavior traces, data-handling boundaries, memory provenance, tool-call timelines, handoff trails, and blocked-decision states.
- Evidence
- Aviation records, system dashboards, and customer diagnostics trained me to treat visibility as an operating requirement.
Aviation records + Apple release validation
Reliability Layer
I have worked in environments where accuracy, traceability, discretion, and release discipline matter: aviation technical operations, aircraft records, and Apple release validation.
- Operating standard
- Accuracy, traceability, stakeholder communication, and discrepancy resolution are daily requirements in aviation technical operations.
- Technical focus
- AMOS maintenance data accuracy, FAA-oriented records workflows, release gates, diagnostics, production issues, and escalation paths.
- Agent lesson
- Production agents need reliability standards, audit records, approvals, rollback thinking, and clear ownership when behavior drifts.
Angel-Gaming operations
Distributed Systems
Angel-Gaming is the high-signal experience: global servers, a large user base, security work, staff coordination, delivery ownership, and years of operational support.
- System operated
- A global Minecraft network with servers, staff coordination, community operations, security practices, incident response, and uptime ownership.
- Scale
- 20,000+ active users, eight global servers, 20+ staff members, and coordination across seven time zones.
- Agent lesson
- Agentic systems are also distributed systems: routing, permissions, observability, delivery commitments, failure recovery, and human coordination.
Agentic AI Engineer narrative
Role Fit
I bring the full loop: translate business workflows, build the agent system, operate the production surface, evaluate behavior, and keep humans in control when autonomy needs judgment.
- Builder
- Programming foundation across Python, JavaScript, TypeScript, HTML, CSS, Java, PHP, SQL, backend workflows, and service-oriented thinking.
- Operator
- Production-minded experience with aviation systems, customer diagnostics, records, issue queues, support workflows, and stakeholder communication.
- Manager
- Hands-on leadership through staff coordination, training, security, roadmaps, open communication, and global community operations.
Resume evidence trail
Resume Ledger
A compact ledger of the roles and systems that shaped my approach to agentic AI: operational ownership, validation discipline, traceable records, and people-first technical support.
Current target: AI agent engineering roles where production agents, evaluation, governance, operations, and human oversight all matter.