AI Architect - Remote

What We’re About

At CentralSquare, we don’t just build software - we power public servants and uplift communities with Hero-Grade Technology. Every line of code, every feature we deliver helps heroes across North America protect, serve, and save lives. When you join us, you become part of a mission-driven team creating technology that makes communities safer and stronger.

Your Growth Matters. We believe heroes deserve opportunities to rise. That’s why we invest in your career with mentorship, learning programs, and clear paths for advancement. If you’re motivated, there’s no limit to how far you can go.

Your Commitment Deserves Reward. We offer competitive compensation and a benefits package designed to support your life inside and outside of work—tuition reimbursement, parental leave, paid volunteer hours, and unlimited PTO. Plus, our flexible work environment gives you the freedom to balance your heroic work with personal well-being, whether you’re in the office or remote.

Join us and help build the tools that power real-life heroes. Together, we make a difference.

The Opportunity

This is an individual-contributor architecture role, and a deliberately hands-on one. You will spend most of your time designing and building — prototyping, writing production code and infrastructure, and working directly in the codebases involved. You will also spend meaningful time with product teams across the company, each with its own architects, roadmap, and deep domain expertise, where your influence will come from collaboration and demonstrated results rather than from reporting lines. 

Expect variety. A given week may include architecture and design work, implementation, embedding with a product team to resolve a dependency, and presenting ideas to internal stakeholders and to customer design partners. 

What You’ll Do

Design and build agent systems. Core reasoning loops, planning, state and memory management, tool interfaces, sub-agent delegation, and orchestration for multi-turn and autonomous agents. 

Make agent output trustworthy enough to act on. This is central to the role rather than a final compliance step: 

• Confidence scoring and routing — determining what can proceed automatically, what warrants a brief human review, and what requires a person to handle directly 
• Human-in-the-loop design that keeps professional judgment, and accountability, with the practitioner 
• Audit trails and explainability sufficient to withstand evidentiary and public-records scrutiny 
• Evaluation practice: golden datasets, regression gates on model and prompt changes, staged and canary rollout, comparison of candidate models against those in production, and monitoring for drift after release 
• Adversarial and failure-mode design: prompt injection and jailbreak threat modeling, containment of the effects of an incorrect action, deterministic controls around probabilistic components, and clear decisions about capabilities we choose not to build 
• Agent-level observability from the outset — which model call, which tool, which inputs, and the basis for the agent's decision 

Make the existing portfolio accessible to agents. Our products span a wide range of technologies, from modern .NET and Python services to long-lived platforms that predate current frameworks. You will build MCP servers, APIs, wrappers, and adapters that expose product capability to agents; contribute to the data pipelines and ingestion layers that feed them; and help instrument existing applications to understand how they are used today. 

Partner across product teams. You will work alongside architects and engineering leaders who know their products deeply. Depending on the situation, that may mean designing a solution with a team and supporting them as they build it, or coming up to speed on their stack and delivering the work yourself — then walking the team through the approach so they can carry it forward. 

Engage with customers. Work with our design partner agencies and end users to ground the work in real operational practice, and bring what you learn back into the design. 

Model current AI-assisted engineering practice. You will bring a well-developed harness of your own: skills and commands, hooks, MCP servers, sub-agents, specification-driven workflow, and deterministic quality gates — compilers, test suites, static analysis, custom lint rules, and architecture-conformance checks — so that non-deterministic tooling produces predictable, standards-conforming output. Teams you work with will learn a great deal simply from how you work.