Why we exist
Cities still run permitting on PDFs, a rigid patchwork of systems, tribal knowledge, and long email chains. The result is bottlenecks, frustrated applicants, and not enough homes, or anything for that matter, getting built. The country is roughly 5 million homes short, and permitting delay is one of the biggest brakes on closing that gap: UCLA found that a 25% reduction in approval time increases housing production by a full 24.6%. Fewer review cycles convert almost directly into more homes. Govstream.ai exists to pull that lever. Our mission is to make building easier, cheaper, and much faster by turning permitting into one continuous, conversation-driven workflow, guiding every step from the first zoning inquiry through final inspection. The North Star is same-day construction permits, down from the 9 to 12 months a review can take today.
What we're building
Permitting is linear but complex, spanning multiple departments and cycles of review. Our bet is that AI can turn it into one adaptive, conversational workflow that draws on both the applicant's inputs (conversation and docs) and the city's data (zoning codes, GIS, historic permits, reviewer notes) to push every step forward. That means interpreting codes, reasoning over architectural plans, managing long-running workflows across web and email, and orchestrating AI and legacy systems into accurate, traceable, cited guidance. We operate on both sides of the counter: real-time guidance for the builders applying for permits, and decision support for the staff who review them.
We've mapped products directly to that process:
- PermitGuide: answers pre-application zoning and permitting questions across web, email, and phone, in any language, always with citations, and surfaces what the community is asking.
- Application Assistant: guides an applicant from the early research stage, where they figure out what's allowed and what's required, through assembling a clean, staff-ready submission, into a first review that helps staff triage and clear backlog roughly 2x faster. Built on an events-based architecture that pairs deterministic checks with inference loops.
- Plan Review (next): reads and interprets the plans, automates deeper code checks, learns from reviewer comments, and keeps pushing toward a Yes.
These are the beginning, not the destination. We're building toward an end-to-end, AI-native permitting system that spans the full lifecycle, from the first zoning question through submittal, review, inspections, code enforcement, and payments.
Just as important as the products is how we make them scale across very different jurisdictions. We're investing heavily in automating customer onboarding, and in a recursive loop that turns a city's annotated historical permits into configuration. Each cycle teaches the system that jurisdiction's real rules and improves how well our configuration guides an applicant to a complete, review-ready application on the first try, which in turn produces better data to learn from. It's the loop that lets us bring a new city online quickly and keep getting sharper after we do.
The role
We're building an exceptional engineering team, and the core technical problem is the reason. We're building a conversational AI system that reasons over city codes (text), architectural blueprints (visual and spatial), and GIS data (geospatial), adapting to every jurisdiction's unique rules. It's not a chatbot. It's an agent-based workflow with deep memory, complex orchestration, citations, and full traceability. This isn't an LLM wrapper. The work spans the stack, and every layer exists to serve that one problem. As a Software Engineer, you will own and build across all of them:
- Core Systems & Infrastructure: the orchestration backbone for long-running, stateful workflows and the integrations into zoning codes, GIS, and legacy permitting platforms. Elixir and the BEAM are our backbone here, chosen for exactly this kind of concurrent, stateful work. We hire for strong fundamentals, not a specific language, so you don't need Elixir on day one.
- AI + Data Engineering: the reasoning layer, reliable and explainable services that interpret plans, codes, and emails and deliver traceable, cited outputs. This is applied AI that we ship and run in production. We have deep respect for research, and the engineers we're looking for love putting it to work in front of real users, owning the path from model to production, not just the model.
- Conversational Interfaces: the surface that makes a complex, agentic workflow feel intuitive for applicants, reviewers, and staff.
- Product Velocity: pairing across backend, AI, and frontend to ship quickly, learn from real users, and iterate with cities in production.
- Outsized Impact: helping define the engineering standards, architecture, and culture that this problem demands, alongside the rest of the team.
What you'll bring
- A sound CS and software engineering foundation.
- Deep curiosity, self-direction, and intrinsically high standards.
- You've designed, built, and shipped SaaS and cloud distributed systems that run in production.
- Startup-tested instincts: you love 0-to-1 greenfield work, but you have also scaled what worked.
- Full-stack range and comfort learning new tools, languages, and ways of working. Experience with the Elixir and BEAM ecosystem (OTP, Phoenix), our backbone for concurrent, stateful work, is a bonus, not a requirement.
- Fluency in agentic systems: comfortable building and tuning the harness that lets a model plan, call tools, and manage long-running, stateful work; rigorous about evals so quality is measured rather than guessed; and at home with context and memory engineering, tool design, and the deterministic guardrails that keep non-deterministic models reliable in production.
- The ability to actually code, deeply, and to be the boss of the AI. We use AI coding agents prolifically, and we hire engineers who direct them, review their output, and own the result.
- A natural pull toward customer outcomes and user experience.
- A track record of having built something objectively remarkable.
- And, simply, a good human being to work with.
How we work
We're a small, high-performance team: transparent, fast, ownership-driven, and deeply focused on creating value for cities and residents. Everyone builds and ships. We work in the open, collaborate closely, and review each other's PRs, and what we optimize for above all is ownership and speed. We're heavy users of AI, Claude in particular, but humans still own every review, and we genuinely value the craft of coding. We avoid bureaucracy and waste and pursue excellence in everything we do.
Traction
We're live in Bellevue and Louisville, with active pilots in 8 cities, and we're well funded and at an inflection point in our growth. By Q4 2026 and into Q1 2027, we expect to have 15-20 cities on the platform, positioning us for a strong Series A. Our ambition is to solve permitting across thousands of cities, not just in the US but globally. We're early enough that every engineer on the team is shaping the architecture and the culture.
Details
- Location: Remote-first, but US-based only. Greater Seattle Area or Berkeley, CA preferred, not required.
- Compensation: Competitive salary and meaningful early-stage equity.
- How to apply: Write to careers@govstream.ai with a short note on something you've built that you're proud of, and a link to it if you can share one.