About
Building agentic systems since 2018. Engineering for the Navy since 2019.
The day job
Engineering for the Navy, since 2019.
I have been a Computer Scientist at the Naval Undersea Warfare Center since 2019, writing software for Navy research and development programs: analysis, prototyping and testing, largely in Python and C++ against real-time systems with hard measurement requirements.
Much of that work was machine learning applied to sensor and signal data — classification, clustering, dimensionality reduction and anomaly detection, for signal analysis and automated decision support. I guided a team of Navy engineers on a machine-learning data-analytics application that we demonstrated to government stakeholders.
I hold a B.S. in Computer Science from the University of Rhode Island, earned in 2022. That grounding is the reason I treat evaluation as part of engineering rather than an afterthought.
The other thread
Independent work, since 2018.
I have been building on my own time since 2018, starting with an AI robotic canine that ran its models on-device, and arriving at the systems I work on now: openagent-code, a self-hosted autonomous coding agent; OpenAgent, a five-service platform for running agents; CareerAgent, a fourteen-service product specialised from that same architecture; and BoeNet, conditional-compute research written from scratch in PyTorch. I also maintain Arcus Code, a hardened distribution of a 266,000-line open-source coding agent.
The thread running through all of it is the unglamorous half — the permission engine rather than the orchestration loop, the evaluation harness rather than the demo. I build the part that decides whether a system can be trusted to run without someone watching it.
Each project is written up in detail on the projects page, including what I would do differently.
Where they meet
Owning the AI roadmap, from 2026.
Since 2026 I have worked with senior leadership on the organization’s AI roadmap: identifying where AI genuinely applies, assessing feasibility, helping set technical direction, and developing the briefings that translate AI capability into concrete applications for leadership. I also build the codebase for those applications, so the recommendation and the working software come from the same hands.
The independent work is what led to the role I hold now. It is the part of my record I trust most, because it was evaluated by an employer and acted on rather than simply asserted on a page like this one.
How I think about this
What I believe after building these systems.
- 01
The permission engine matters more than the orchestration loop.
Deciding when a human belongs in the loop is what separates an agent people trust with real work from one they quietly stop using.
- 02
The dataset and the evaluation decide whether it works.
Far more often than the model does. I write the eval and the verifies before I tune anything, so improvement is something I can measure rather than something I can feel.
- 03
Build from first principles, then reach for a framework.
I write the transformer, the router, the loop by hand at least once, because you cannot debug what you do not understand.
- 04
Own it end to end.
Architecture, data model, deployment, monitoring. The parts nobody owns are the parts that fail in production.
- 05
Document honestly, including the limitations.
I write for someone who did not build the system, and I say plainly where it stops working.
Capabilities
What I work with.
Agents & LLM applications
- Orchestration loops, tool and function calling, subagents
- Cross-session memory and context compaction
- Retrieval-augmented generation on pgvector
- Permission engines and human-in-the-loop approval
- Evaluation harnesses, LoRA-SFT distillation
- MCP integration
Platform & infrastructure
- FastAPI, REST APIs, microservices, SSE streaming
- Provider-agnostic model routing with failover
- vLLM, Amazon Bedrock, RunPod GPU
- PostgreSQL, pgvector, Docker, CI/CD, AWS
- Observability for latency, token cost and system health
Engineering & ML
- Python, C++, Go, TypeScript, SQL
- PyTorch, TensorFlow, scikit-learn
- Supervised and unsupervised methods on signal data
- Conditional-compute model development
- Real-time and resource-constrained systems
- React, Next.js
Credentials
- Citizenship
- U.S. Citizen
- Education
- B.S. Computer Science · University of Rhode Island · 2022
- Based in
- West Warwick, RI · Remote