Agent routing, memory, tool access, approvals, and operator visibility.
Agentic AI engineer | airline Tech Ops | release validation
I build agent systems people can operate.
I am not interested in agents that look impressive for five minutes and then hide what they did. My work is about the pieces that make an AI system usable after the demo: the workflow, the record it searched, the tool it called, the test it passed, and the moment it hands work back to a person.
Test runs before autonomy expands.
Traces for memory, tools, handoffs, and blocked decisions.
Records, release validation, diagnostics, and escalation work.
airline technical operations
aircraft records and AMOS
10+ Apple OTA validations
20,000+ users supported
AgentReef and Angel Software direction
A control room for agents, not a magic box.
AgentReef is where I am working through the question every serious agent project runs into: who decides what the agent can do, what context it is allowed to use, and when a human needs to take over?
The answer is not one giant prompt. It is a system: routing, retrieval, scoped tools, memory, approvals, traces, and review points that people can actually inspect.
What this proves
- Builder
- I can turn an agent idea into interfaces, flows, and working system boundaries.
- Operator
- I think about what breaks after launch: access, drift, missing context, bad handoffs, and unclear ownership.
- Manager
- I care about the person using the system as much as the model behind it.
route maintenance-record question detected
context source set narrowed before answer
tool write action blocked until approved
human review requested for judgment call
Resume evidence
The resume evidence is the point.
I am aiming at agentic AI work because my background is oddly useful for it. I have worked with aircraft records, release validation, technical support, distributed communities, and the unglamorous parts of systems people rely on.
Tech Ops systems work where accuracy, access, and operational handoffs matter.
Aircraft records, AMOS maintenance data, FAA-oriented traceability, and discrepancy resolution.
Diagnostics, repair workflows, escalation judgment, and high-volume customer issue resolution.
10+ OTA updates tested with manual checks, automation, bug reporting, and release review.
20,000+ users, eight global servers, 20+ staff, security practices, and uptime ownership.
AI Battle Arena
scenario run
Records lookup
pass
uses cited source
Tool request
hold
approval required
Ambiguous issue
handoff
human judgment
Bad context
fail
blocked before action
Arena: evaluation before autonomy
Before an agent gets freedom, it should fail safely in rehearsal.
Arena is my evaluation project. I do not want to measure agents only by whether they sound right. I want scenarios, regressions, tool-use checks, grounding checks, latency, cost, and failure notes.
Apple release validation shaped this thinking. A release does not become ready because it feels promising. It becomes ready because it survives review.
What this proves
- Validation
- I have worked inside test and release routines where repeatability matters.
- Agent lesson
- Autonomy should expand only after the agent proves where it succeeds, where it fails, and how it recovers.
Aquarium: behavior observability
I want agent behavior visible before people trust it.
Aquarium is about watching the system work. What did the agent retrieve? Which memory did it use? What tool did it ask for? What did it refuse to do? Where did a person step in?
That matters because operators do not need a mystical answer. They need a trail they can inspect when something looks wrong.
retrieval traces
tool-call timelines
blocked decisions
human handoffs
behavior trace
source checked
handoff ready
memory
tool
review
Aviation records + release validation
Aviation and release work changed how I think about AI.
Aircraft records taught me that the boring parts are often the important parts: clean data, a traceable source, a complete handoff, and a way to resolve discrepancies without guessing.
Release validation taught me the same lesson from another angle. If people are going to rely on a system, the system needs checks before it gets more responsibility.
FAA-oriented records work
AMOS maintenance data
Apple OTA validation
bug reporting and review gates
What I am looking for now
I want to help build agent systems people can actually run.
The roles that interest me sit between AI engineering, workflow design, operations, and evaluation. Airlines are an obvious fit because I know the records and Tech Ops side, but the pattern also applies anywhere people need AI to work inside real process.
I can help a team find a first useful agent project, build the workflow, test the behavior, make the work visible, and keep the human decision points clear.