The posting
Havenpark owns and operates manufactured housing communities across several states, supported by a lean central team. Selling agents, community managers, accountants, and marketers all do repetitive work every day across many communities, and AI gives us that time back. It already runs in production in parts of our sales, marketing, and finance work, and this role is how it reaches the rest of the business.
You will sit with the people doing the work, learn how their day actually runs, and then build and operate the AI companions, agents, and automations that change it. The foundation is in place: governed data in Snowflake, a semantic layer, and a modern Microsoft and Azure environment. This is an internal engineering role rather than a customer-facing one, with direct access to the executives who sponsor the work and the users who live with it, and what you ship gets measured in hours returned, adoption, and reliability.
Building AI solutions
- Assess how work is actually performed, role by role, and prioritize where AI removes steps rather than only speeding them up, weighing hours returned, error reduction, and how many people the change reaches.
- Design, build, and ship AI companions, agents, and workflow automations from prototype through production, for users including selling agents, community teams, accounting, and marketing.
- Integrate with the systems where work happens: HubSpot, Microsoft 365, Snowflake, our property management and finance platforms, and vendor APIs. Choose the runtime that fits the workload rather than forcing everything into one tool.
- Build on the governed data models and semantic layer the analytics team maintains. You consume them rather than own them, and you tell us what agents need.
- Maintain reusable patterns, prompts, tools, and evaluation sets so the third solution is faster to build than the first.
Operating what you build
- Own evaluation, monitoring, exception handling, and escalation for every production workflow. Define what good looks like before launch and know within hours when something drifts.
- Enforce human review for consequential actions. Anything touching a customer, a resident, money, or a record of record keeps a person in the loop until the evidence says otherwise.
- Manage model usage and cost, ship through source control and pull requests, and document each solution well enough that someone else could support it.
- Keep the inventory of agents, prompts, and integrations current, including who owns each one and what it can access.
Working with the business
- Partner with function leaders to pick the right problems, agree on what success means, and confirm the value was realized after launch rather than at the demo.
- Embed with users while a solution is new. Watch how they use it, find where it fails or gets ignored, and iterate. Adoption is part of the job.
- Train users and their managers on what a solution does, what it does not, and how to escalate when it gets something wrong.
Governance and security
- Work within Havenpark's AI usage policy and bring new tools and use cases through the intake process, involving Legal and HR early when a use case touches applicant, resident, or employee data.
- Give every agent and integration its own identity with least-privilege access. No shared API keys, no over-permissioned service accounts, and no production credentials in code.
- Log what agents did, what data they used, and who approved consequential actions, so any outcome can be traced and explained.
How we work
What you bring
- 5+ years of shipping and supporting software that real users depend on, not prototypes.
- Strong Python and SQL. Backend work is central to this role, and a second language such as TypeScript, Node.js, or Go is welcome.
- Enough frontend ability to put a usable interface on what you build, whether that is React and TypeScript, Streamlit or Power Apps, or an experience embedded in a tool people already use.
- REST APIs, webhooks, and event-driven integrations, and services that connect SaaS platforms reliably, including retries, idempotency, and failure handling.
- Relational data and data modeling, with working knowledge of document or vector stores and caching, and judgment about which fits.
- CI/CD, pull-request-based development, automated testing, and infrastructure as code. Azure is our cloud; AWS or Google Cloud translates.
- Daily use of AI coding tools such as Claude Code, Cursor, Codex, or GitHub Copilot, and a clear sense of where they need supervision.
- Hands-on experience putting large language models into production: prompt design, structured outputs, function and tool calling, and multi-step agent workflows.
- Retrieval-augmented generation, embedding models, vector search, and tool-use standards such as the Model Context Protocol, plus the judgment to know when simple retrieval beats an agent.
- Direct experience with at least one major provider's APIs (Anthropic, OpenAI, Azure OpenAI, or Snowflake Cortex), choosing models and runtimes on fit, latency, cost, and data sensitivity rather than novelty.
- Designing evaluations, monitoring, and guardrails, and defining what good looks like before launch.
- Honest build-versus-buy judgment. You can assess a vendor's AI feature, recommend configuring it when that is the better answer, and say when AI is the wrong tool.
- You have put an AI workflow into a live business process and can describe what broke and what you changed. We will ask to see something you have shipped.
- The ability to sit with a business user, map how their work happens, and see where steps can be removed. You are as comfortable in a sales office as in a code review.
- Least privilege, secrets management, dedicated service identities for agents, and care with applicant, resident, and employee data.
- Clear communication with executives, frontline users, and technical peers, including honest assessments of what AI can and cannot do reliably today.
- Snowflake, including Cortex, and agents built over semantic layers or governed data models.
- Azure Data Factory, Power Automate, and Agent 365.
- Enough dbt familiarity to read data models without owning them.
- Containers and scheduled or long-running cloud workloads.
- Property management, real estate, housing, or another business with distributed field operations.
