The opportunity
You get to build a category-defining tax product from a blank page, in the window where one senior engineer with frontier AI tooling can deliver what used to take a team of five.
Tax is a high-stakes, regulated, globally enormous market that has barely been touched by AI-native software. taxOS is built by founders who know tax cold, with a domain reviewer on the bench and a fractional CTO already in place. The substrate, the calculation engine, the AI orchestration, the practitioner workspace: all of it is yours to architect and ship through to a live filing season.
This is the moment to build the thing that would have taken a venture-backed team eighteen months to assemble, in a fraction of the time, with your name on it.
The autonomy
You are not joining to execute someone else’s plan. You make the substrate choices. You set the architecture. You own CI/CD, observability, security posture, and the regression discipline a tax product demands.
You will push back on the founders when the product direction is wrong, and commit fully once a decision is made. They are tax experts, not engineers, and they know it. Your judgment is why we are hiring you.
We have a preferred stack documented in the PRD: TypeScript / NestJS, PostgreSQL with pgvector, React, Anthropic Claude as primary LLM, AWS or Cloudflare for substrate. Nothing is locked in. If you arrive in the first month with a strong reason to change a choice, we want to hear it. Two non-negotiables: it has to be a modern web and cloud stack, and it has to be mainstream enough that we can hire the second and third engineer who already know it.
For the first year the team is deliberately small and trust-based: the two founders, you, and a second engineer, plus a third who joins around month 3/4 to push UI velocity. No hierarchy to navigate. No committee to convince. You build the system.
What you will actually do
- Architect the platform and own the substrate choices.
- Design the ledger. taxOS needs a tax-aware data model that captures entity relationships, source documents, and the time dimension tax demands (transaction date, filing period, amendment history, audit trail). Graph, event-sourced, bitemporal, or some combination: that call is yours to make and defend.
- Build the deterministic calculation engine and the agentic orchestration layer that surrounds it. The orchestrator decomposes practitioner intent into deterministic calculation steps and probabilistic classification steps, with explicit guardrails, retries, and human-in-the-loop checkpoints between them. Single-prompt chains are not the bar.
- Ship document ingestion, classification, and the practitioner workspace through to a live filing season.
- Stand up CI/CD, observability, security posture, and the regression discipline a tax product demands. A 0.5% silent calculation error is a serious failure, and the system has to make that class of error impossible.
- Walk a practitioner through the product in two minutes. Selling it to a senior accountant is part of the job.
What “founding engineer” means here
You write the code. We are not hiring a CTO, an architect-on-paper, or a manager. The product gets built by the person we hire, with judgment, taste, and the 2026 toolchain.
If you drifted into management, architecture, or fractional work because pure coding stopped being interesting, and you have come back to building because AI tooling has changed what one person can deliver, this seat was built for you.
What we care about underneath the stack
- Production experience with frontier LLMs. Shipped systems, evaluation harnesses, a real cost-per-call number you can defend. Not prototypes.
- Agentic orchestration in production. Multi-step agent workflows with tool use, retries, fallbacks, and human-in-the-loop checkpoints. You have a view on when to constrain an agent and when to let it reason.
- Graph or event-based data modelling. Comfortable with graph databases, event-sourced ledgers, or bitemporal data, with a view on when each is the right tool. Tax is fundamentally relational and temporal, and the ledger has to reflect that.
- A track record in a regulated or high-stakes domain. Fintech, healthtech, tax, payments, legal tech. You have internalised that probabilistic outputs need deterministic guardrails and that audit trails matter from day one.
- Strong on either Python or TypeScript, willing to use whichever the project needs.
- Comfortable owning the full stack: backend, data model, frontend, infra. The job is not divisible into roles.
Who you are
A few things matter more than years of experience.
- You are excited about what AI changes for builders, not threatened by it. The right person sees Claude and similar tools as a lever, not a risk.
- You are a strong communicator. You will translate the founders’ vision into a working system, and translate back when something they want will not work. That conversation has to be crisp.
- You overlap Europe or US East Coast working hours comfortably. Async is fine once the product is built. While you are still figuring out what you are building, same-ish hours matter.
- You are senior enough to be opinionated and humble enough to be wrong. Both are the job.
- You do not need a tax background. Domain expertise comes from the founders and the senior practitioner.
Engagement
- Full-time, fully remote within Europe or US East Coast.
- Twelve-month minimum commitment. The expectation on both sides is that this becomes long-term.
- Competitive base salary, benchmarked to senior engineering in your region.
- Founding-engineer equity: 1.5 to 3.0 percent post-seed, four-year vest, one-year cliff.
- Mutual NDA and standard IP-assignment terms before any confidential material is shared.
How to apply
Apply below. Include:
- One or two production systems you have personally built that are comparable in complexity. What it did, what you owned, what you would do differently today.
- Links to code, writing, or anything that shows your work. A real artifact beats a polished CV.
- Your location, notice period, and target base salary.
- One sentence on why taxOS specifically. Not why startups. Why this one.
We reply to every serious application.