AI Engineering · U.S.-based remote
AI Operations Data Engineer
Reports to: VP of Business Development & Partnerships
Full-time · Remote, with a hybrid option in the San Francisco/Sacramento area · U.S. time zones
About Shovels
Shovels turns building permit data into intelligence that customers use to make decisions about real estate, construction, insurance, data-center siting, and more. We bring records from local governments into a platform that helps customers understand what is happening in the built world.
AI is already part of how we build and operate. You'll join experienced, enthusiastic colleagues who try new tools early and put them to work. We're looking for someone who loves that experimentation and wants to make the best ideas dependable enough for the whole company to use.
The opportunity
We collect records from local governments with a fleet of AI agents. You'll make that fleet run like production software. Reporting to the VP of Business Development & Partnerships, you'll build and own those agents end to end — the harnesses, evaluations, and automated workflows they run on, and the adapters that carry what they collect into the Shovels platform.
You'll work with our jurisdiction partnerships team — the people who build Shovels' direct relationships with city and county governments. They establish those partnerships through phone calls, emails, and in-person meetings. Our partners then send us their records through public records requests, web portals, emails, and forms, which brings updated records into the Shovels platform every month.
A fleet of AI agents already runs much of this — triaging incoming mail, validating the files that arrive, keeping our CRM current, and filing requests across 1,000 jurisdictions and counting. It works, and it needs to be production grade. You'll learn the fleet alongside the team and then harden it: agents that do their work predictably, evaluations that catch quality regressions before a person does, and the freedom to change models without breaking anything. You'll do this without interrupting the monthly collection the team already depends on.
Over the next two years this work expands well beyond building permits — right-of-way permits, as-builts, business licenses, liens, and deeds — across several times as many jurisdictions. The platform you build is what makes that possible.
You'll also own what happens after collection. Today, records arrive and the work of getting them into the platform queues behind other engineering priorities. You'll write and maintain the adapters that bring each jurisdiction's records into production, so new data types land without a backlog forming behind them.
This is a hands-on engineering role, held to the same bar as the rest of our engineers. What's different is where your priorities come from: the partnerships team sets them, not the engineering roadmap. You'll write code that the engineering team reviews, sit in the engineering channel and the weekly engineering calls, and onboard with both teams — so you keep growing as an engineer while working on problems engineering isn't staffed to take on. You own what you build after it runs. You break down problems, test assumptions, diagnose failures, and make systems more reliable.
What you'll own
Make the agent fleet production grade
- Partner with the jurisdiction partnerships team to increase collection capacity, broaden coverage across local agencies, and support new data types.
- Build and run the agents so they complete their work predictably. Investigate failed or stalled tasks, fix the cause in the agent or its harness, and add a check so the same failure does not return.
- Build evaluations on real collection work, and pick the right model for each agent based on what that agent actually does — not every job needs the same one. Use those evaluations to change models without breaking the work.
- Maintain agent instructions, configurations, schedules, and operating documentation. Make output-quality checks, human review, and points of intervention clear.
- Keep one prioritized improvement backlog, with the team's core operating needs first.
Get jurisdiction records into production
- Write and maintain the adapters that bring each jurisdiction's records into the Shovels platform, and own them through to production rather than handing them off.
- Work in our main engineering repository to the team's standards, with your changes reviewed by the engineering team.
- Support new data types as the team takes them on, so coverage grows without a queue forming behind it.
- Help build what we give jurisdictions back, such as the dashboards that show a city what its own permit data says.
Make sound technical and operating decisions
- Weigh quality, reliability, speed, and total cost — including retries, failures, and human review or rework — and explain the tradeoff you made.
- Choose the simplest implementation that meets the need. Work with engineering to decide when to build an agent harness (software that coordinates an agent's tools and execution), use an existing one, or use no harness. Build and maintain the harnesses you choose. Keep unnecessary infrastructure and third-party dependencies out.
- Build, deploy, and troubleshoot workflows in Amazon Web Services (AWS) and OpenClaw, a tool for running AI agents. Define ownership and escalation paths, and review changes to agents and their permissions.
- Work with the infrastructure team on secure access and the SOC 2 requirements relevant to agent operations. Protect credentials and data.
- Be clear about what works reliably, what needs human review, and what is still experimental.
Share what you learn
- Document the fleet so other people can operate and improve it without relying on one expert.
- Bring useful practices back to the rest of the company as the same patterns show up in other teams' work.
What you'll bring
- A bachelor's degree.
- At least 3 years of work experience. This is not an entry-level role.
- Fluency in English.
- Curiosity that turns into useful work. You enjoy building with AI, follow new tools, and can show how your experiments have helped other people.
- Hands-on AI engineering experience. You have built and run agent or automation systems in production that other people use every day. You are proficient in AWS, including configuration, logs, and troubleshooting. Experience with OpenClaw or a similar agent runtime is a plus.
- Software engineering skills. You write production Python and can show working systems you built. You have worked on data pipelines or integrations where someone else depended on your output being correct. You break down unfamiliar problems, test assumptions, and find the cause of failures.
- Good systems judgment. You have built evaluations on real tasks and used them to choose models and harnesses. You can explain tradeoffs among quality, reliability, speed, cost, and complexity.
- Strong security judgment. You understand permissions, credentials, data access, and where human review belongs.
- Follow-through and clear priorities. You document repeatable processes, handle competing requests, and close the loop on recurring problems.
- Comfort working on your own track. You'll be the engineer embedded in a team of non-engineers. You hold your own standards, seek out review rather than waiting for it, and don't need a roadmap handed to you to know what matters.
- Warm, practical communication. You enjoy helping technical and nontechnical colleagues learn, welcome questions, and make unfamiliar tools feel manageable.
Experience with government or public-sector data, document and file processing at scale, or internal business-process automation would be useful. So would working across model providers, running scheduled automation, or partnering with infrastructure and security teams.
What success could look like
- Our jurisdiction partnerships team collects more data across more jurisdictions, with fewer recurring failures and less manual supervision. The agents keep their quality when we change models, and the evaluations show it.
- New data types reach production on the partnerships team's schedule rather than waiting on the engineering roadmap, and the adapters behind them have an owner.
- The team can see agent health, output, costs, and configuration changes, and uses that evidence to improve quality and total operating cost.
- The monthly collection the business depends on never goes dark while any of this happens.
Interested?
Interested in building the future of construction intelligence with us? Apply with your resume and a short note about your experience.
Need an ADA accommodation, have a question or hit a tech barrier? Reach out directly to [email protected].