The Role
We're hiring a Staff Software Engineer, Applied AI. You'll set the technical strategy for how Ruxlo builds AI into the product, from a single well-designed LLM call to long-running autonomous agents, and you'll be the technical anchor for applied AI across teams.
The range is the point. One week you're tuning natural language search over property records or extracting structured data from a scanned receipt. The next you're designing an agent that runs in the background for hours, calls dozens of tools, pauses for a human approval, and resumes exactly where it left off after a failure.
Some of these agents act on a user's behalf. A homeowner might ask an agent to request quotes from contractors their neighbors used, file a receipt and warranty, or follow up until a contractor confirms a job. A contractor might hand an agent the import of their work history. These agents carry a user's authority, so they ask before anything consequential, stay inside what that user is allowed to do, and leave a record the user can review.
You know when a prompt is enough, when retrieval is needed, and when only an agent will do. The engineer who succeeds here ships all three to production, measures them honestly, keeps their cost in check, and leaves behind a platform other engineers build on.
What's hard: The patterns for production agents are still being invented, and you'll be setting direction without a playbook to lean on. If that sounds like the appeal rather than the warning, we should talk.
What You'll Own
- Own the architecture of Ruxlo's AI systems end to end: model access and routing, retrieval, prompt and tool layers, agent runtime, evaluation and cost
- Ship LLM-powered product features, including natural language search, summarization, structured extraction and classification over property, ownership and contractor records.
- Ship autonomous background agents that verify, reconcile and enrich records and keep them fresh.
- Ship agents that act on a user's behalf, such as requesting quotes, following up with contractors, importing work history and filing documents, with the user's consent at each consequential step.
- Design durable, long-running agent workflows on a durable-execution engine such as Temporal, with retries, timeouts, idempotent tool calls and safe resume after failure.
- Stand up evaluation for every AI feature: golden datasets, offline evaluation, regression gates in continuous integration and production sampling.
- Set the standards for safe AI use, including prompt-injection defense, least-privilege tool access, delegated authority scoped to what the user could do themselves, and audit trails the user can review.
- Make sound build-versus-buy calls across model providers, frameworks and vector stores, and treat cost and latency as first-class constraints alongside
- Build shared AI tooling, skills and Model Context Protocol servers, and mentor engineers on building with AI.
What You Bring
Core Competencies
- Excellent verbal and written communication skills
- Ability to work well independently or as part of a team
- Adaptable and comfortable working in ambiguity
- Strong ownership and follow-through
- Effective use of AI tools in daily work
- Sound decision-making and prioritization
- Open to feedback, and able to give it well
- Collaborative across functions
Position Specific Requirements
- 8+ years of engineering experience, including a technical lead role on systems other engineers built on.
- LLM-powered features you shipped to real users, such as retrieval-augmented search, extraction or classification. You can walk us through what worked, what failed and how you measured it.
- At least one autonomous, non-conversational agent you took to production: triggered by events, schedules or data, and completing multi-step work without a person in a chat window.
- Agents you built that act for a user in other systems, including delegated permissions, consent and confirmation flows, and undo.
- Hands-on experience with durable execution or workflow Temporal is strongly preferred; Restate, Inngest, AWS Step Functions or similar also count.
- Deep command of prompt design, structured outputs, tool use, Model Context Protocol, embeddings and vector search, and context-window management.
- A track record of evaluation and observability for AI in production: tracing, cost metering and quality measurement without a human grading every
- Strong backend fundamentals in Python or TypeScript, PostgreSQL, and distributed systems on AWS with Docker and Kubernetes.
Preferred Experience
- Agent frameworks and runtimes such as the OpenAI Agents SDK, Claude Agent SDK, LangGraph or OpenClaw.
- Classic machine learning for entity resolution or record matching over large structured or geospatial datasets.
- Fine-tuning or distilling smaller models for high-volume, narrow
- Cost engineering at scale: model selection per task, prompt caching, batching and token budgets.
- Open-source contributions to agent runtimes, workflow engines or Model Context Protocol servers.
- Experience at a founder-led or bootstrapped startup
This is a dedicated, in-house role, we are not considering fractional, part-time or consulting arrangements. Candidates must be authorized to work in the United States, we do not offer visa sponsorship. We do not accept candidate submissions from agencies or third-party recruiters for this role. Unsolicited resumes sent to Ruxlo or any of our employees will be treated as direct applications and no fee will be paid for any resulting hire.
