Scratch Engineer (Full-Stack AI, Agentic Systems)

The mission, and the wedge

US commercial real estate is a $26 trillion asset class, and the financial brain that runs it is a patchwork of 1990s ERPs, email, and Excel.

We’re building its replacement: the intelligence layer that will eventually power how every property fund accounts, reports, and decides.

We’re starting where the pain is sharpest: the close.

Real estate funds managing billions run their books across property ERPs (Yardi, MRI, AppFolio, JDE, RealPage) and close them in Excel, by hand, every month.

A single misplaced cent can blow up an audit.

We automate that close: AI agents that extract, draft, and reconcile, checked by a deterministic engine that proves every number to the cent, with an audit trail back to source.

The agents propose.

The engine proves.

It’s $25B+ of annual finance-and-accounting spend that modern AI hasn’t touched, and we have early enterprise engagements pulling us forward.

Nobody has built this before, which means there’s no playbook, no Stack Overflow answer, no senior person to ask.

That’s the job.

Why “Scratch Engineer”

Because that’s the actual work.

You’re joining a small team of builders where everything is still version one: the schema, the accounting engine, the agent architecture, the deployment pipeline.

Nothing is legacy.

Nothing is decided that can’t be revisited.

The systems you write in your first months will be the foundation that billions of dollars eventually flow through, and you’ll live with your own decisions, the brilliant ones and the ones you’ll rewrite at 11pm.

If you’ve spent your career inheriting other people’s architecture and wondering what you’d do differently from a blank page: this is the blank page.

What this actually looks like

• Week one, you ship.

Greenfield codebase, real problems, no committee.

Your code doesn’t get reviewed into oblivion.

It gets used.

• You’ll make architecture calls with incomplete information, and some of them will be wrong. We care that you notice fast and fix faster.

• Some weeks you’ll design a clean abstraction for waterfall calculations. Other weeks you’ll write a gloriously ugly script at 11pm because a demo is tomorrow.

• You’ll use AI as leverage everywhere: Cursor, Claude, whatever shipped last Tuesday.

If you’re not 3x faster than you were two years ago, this isn’t the place.

Here’s the interesting part about our customers

They won’t tell you what to build.

Ask a fund accountant about their pain points and they’ll say Excel works fine, because they’ve spent ten years getting fast at a process that shouldn’t exist.

The pain is invisible to the people living in it.

So discovery here doesn’t look like user interviews.

It looks like watching a highly paid professional copy-paste across three monitors for twenty minutes, seeing what they can’t see, and building the prototype that makes the future undeniable.

When a controller watches your demo and says “there’s no way this handles our waterfall structures,” that objection is your roadmap, and proving them wrong is the most satisfying work in this company.

If you need customers to hand you a spec, this will frustrate you.

If you want your judgment to be the spec, this is the role.

Who thrives here

• You’ve built things end to end: products, side projects, something at a startup where you were the team. We don’t care if that took you two years or ten, or where (or whether) you went to college.

• You know when to architect and when to hack, and you’ve felt the pain of getting that wrong both ways.

• Ambiguity energizes you. “Figure it out” is a complete spec.

• You’re allergic to “that’s not my job.” Frontend, backend, data pipelines, prompts, demos: you go wherever the problem is.

• Precision matters to you. We work with money. “Mostly right” is wrong.

Who doesn’t

• You want a roadmap handed to you each sprint.

• You need consensus before you ship.

• You measure your career in titles and team size.

• You think AI tools are cheating.

Why join

• The blank page.

You’ll author the first version of systems built to move billions, and watch your early decisions compound into the product, for better and worse. Very few engineers ever get this.

• Founders who’ve done it before.

Stanford and Columbia alumni who have founded companies and had multiple exits.

We know what scale looks like and what it takes to get there.

• Compressed learning.

A year here is five years anywhere else.

You’ll touch agentic systems, financial infrastructure, enterprise products, and customer workflows, in your first quarter.

• No layers.

You work directly with founders.

Your opinion on product direction is expected, not tolerated.

• Competitive compensation and flexible time off.

How to apply

Skip the cover letter.

Email careers@relevanse.ai with:

• A link to the thing you’ve built that you’re proudest of (repo, live product, demo, anything we can see or run).

• Three sentences: the hardest bug or design decision behind it.

• Optional, but it gets our attention: tell us one thing in this JD you’d push back on.

We reply to every real application within 48 hours.

Process is fast: one conversation, one working session building something together, decision.