Senior AI Platform Engineer (Agentic SDLC Framework)

Clariti builds community development software for cities and counties, focusing on permitting, licensing, plan review, and workflows that enable local governments to serve their residents. The company is undergoing an AI-native transformation, aiming for a 4x to 6x reduction in implementation time compared to traditional government software projects.

About the Role

The Engineering team at Clariti operates on an Agentic SDLC framework, which includes agent specs, reusable prompts, orchestration templates, and evaluation harnesses. The Applied AI pod is responsible for this framework and is expanding its use beyond engineering into Professional Services and the Clariti AI harness, enabling non-technical employees across Sales, Customer Experience (CX), Finance, People, and Professional Services to use AI safely for real work.

You will serve as the senior engineer on this pod, focusing on DevOps for AI. Unlike traditional DevOps, you won’t ship product features but instead build the infrastructure that accelerates the work of others. This includes the connector layer, evaluation harnesses, orchestration runtime, cost and quality telemetry, and guardrails that ensure AI outputs are trustworthy for government software.

You will collaborate closely with a Technical Product Manager who owns the roadmap. You will determine how it gets built and much of what gets built, as the line between architecture and implementation is yours to define in this small pod.

What You Will Do:

• Build and evolve the Agentic SDLC framework. Design and implement agent workflows, orchestration templates, and reusable components that build pods rely on. Continuously harden existing elements, extend missing functionalities, and ensure the framework remains fast as model capabilities and delivery patterns evolve.
• Build the connector layer. Design, implement, and operate MCP servers and integrations into core systems, enabling agents and non-technical employees to act on real company data with proper permissions. Treat connectors as production software, ensuring they are versioned, tested, monitored, and follow the principle of least privilege.
• Make evaluations the backbone. Develop and maintain evaluation harnesses that automatically score agent outputs. This includes regression suites for prompts and workflows, quality gates in CI/CD pipelines, and scoring infrastructure to determine whether changes to models, prompts, or workflows improve performance. The mantra here is: If we cannot measure it, we cannot scale it.
• Own AI observability and cost telemetry. Instrument token spend, latency, evaluation pass rates, and usage across all production agent workflows. Build dashboards and alerts to transform the perception of AI from being expensive and mysterious into a managed system with clear unit economics per workflow.
• Engineer the guardrails. Implement permissioning, audit trails, versioning, and output controls to ensure agent-assisted work is compliant and defensible in a government context, where artifacts may be presented to a planning commission. Make the safe path the default in code, not just in policy documents.
• Ship enablement infrastructure. Build skill and template libraries, onboarding flows, and self-serve tooling to enable non-technical employees to quickly start producing real work with AI. Establish feedback loops to route usage data back into the platform roadmap.

What You Bring

• 6+ years as a software engineer shipping production systems, with at least 1 to 2 years of experience building LLM-powered or agentic systems that real users depend on, not just prototypes.
• Strong general engineering fundamentals: you are a senior developer first and an AI specialist second. Distributed systems, API design, CI/CD, and cloud infrastructure should be your home territory.
• Hands-on depth in the current agentic stack: agent frameworks and coding agents (e.g., Claude Code, LangGraph), MCP or comparable tool protocols, structured outputs, and evaluation-driven development. You should have strong opinions about context management and be able to defend them with data.
• A platform temperament: measure your success by the throughput of other teams, write documentation that people actually use, and prefer deleting code over defending it.
• Comfort operating with a small blast radius and high autonomy: this is a small pod with a company-wide mandate, not a large team with narrow lanes.

Nice to Have

• Experience in regulated or public-sector software, where auditability and defensibility of outputs are critical.
• Prior experience in DevOps, platform engineering, or internal developer platform ownership, and understanding the difference between building a tool and driving its adoption.
• Experience in instrumenting and optimizing LLM cost and quality at scale, including model routing, caching, prompt compression, and fine-tuning trade-offs.

How Success is Measured

• 90 days: You know the Agentic SDLC framework end-to-end, understand how build pods use it daily, identify its strengths and weaknesses, and are shipping improvements without regressing pod velocity. You take over operational ownership of existing connectors and evaluation harnesses from the founding team, and your first improvement to the framework, scoped with the Pod Lead based on pod needs, is live.
• 12 months: The framework is measurably faster and more reliable than on your first day, with automated evaluation gates on every production agent workflow. Cost and quality telemetry per workflow drive routing and optimization decisions. The connector layer covers core systems and operates like production infrastructure. The Clariti AI harness serves a majority of non-engineering staff weekly, with uptime and support load manageable by a small pod.