Full-Stack Software Engineer - Discovery

Soraban

AI tax workflow platform for accounting firms

$140K - $230K 0.05% - 0.15% Chandler, AZ, US

Job type:
Full-time

Role:
Engineering, Full stack

Experience:

6+ years

Visa: US citizen/visa only

Skills:

JavaScript, Python, React, Ruby on Rails, Type Script

Soraban is building the AI that takes the manual work out of tax preparation. Our platform runs the entire workflow for accounting firms — client intake (Collect), data entry into their tax software (Connect), AI-native return prep (Prepare), and delivery with e-sign and payments (Deliver).

Here’s how we move: our product engineers build the MVP, ~80% of the way there, validated against a clear roadmap.
Your job is to take those MVPs to production: fast, and properly. Reliability, tests, scale, and the kind of UX polish that makes a skeptical CPA trust AI. This is the highest‑leverage engineering work in the company. We need more builders who are great at it.

This is in‑person 5x a week in Chandler, Arizona.

Not ‘product features.’ Real problems, this quarter:

• Production‑harden MVPs across our four products against a roadmap that’s already scoped — you’ll rarely wonder what to build, the bar is how well and how fast.
• Make document pipelines reliable at scale — classification and extraction across W-2s, the 1099 series, multi‑page K‑1 packages, 1098s, 1095‑As — feeding a human‑in‑the‑loop review flow where accuracy is the product.
• Tame AI non‑determinism — the hard, interesting part. Evals, regression tests, confidence/provenance on model outputs, and guardrails so a fallback or a truncated context never quietly ships a wrong return.
• Build backend systems that hold up — messy real‑world PDFs, integrations with Ultra Tax, Lacerte, Drake, and CCH Axcess, and throughput that scales with the busy season.
• Earn CPA trust through small UX decisions — reviewers approve our AI's work side‑by‑side with the source documents. The interface either makes that fast and trustworthy or it doesn’t.

• Week 1–2: Ship a real fix to production. Get deep on the codebase and the AI pipelines.
• By day 30:
  Own a slice of one product/project. Take an MVP feature from ‘works in the demo’ to ‘works for every firm’ — tests, edge cases, error states.
• By day 60:
  Lead the production‑hardening of a roadmap feature end‑to‑end, and measurably improve the reliability of one AI extraction or review flow.
• By day 90:
  Own a product surface, ship meaningful improvements every week, and move a number that matters.

• Backend strength is non‑negotiable. Data modeling, APIs, reliability, performance. Ruby on Rails is a plus, not a requirement — we care that you can build a solid backend, not which framework you’ve used.
• Product and UX judgment is non‑negotiable. You feel it when a flow is clunky and you fix it without being asked. React / React Native experience is a plus; UX sense is the actual requirement.
• You’re AI‑native. You already build with AI tools every day. You know the best practices of AI‑first development — when to let the model draft and when to take the wheel, and you never ship confident‑sounding slop. AI‑generated code gets tested and guardrailed, always.
• Low ego, team‑first. You want the team to win. You’ll happily spend a day hunting a bug because shipping something solid matters more than your line count. No heroes, no territory.
• Fast and careful at the same time. You move quickly against a clear roadmap and still build it properly. Both, not one.
• You’re a builder. You’ve shipped things — ideally things that weren’t assigned to you. Side projects with real users, a product you launched, a startup you started, substantial open‑source work, indie‑hacking revenue. If you build because you can’t help it, you’ll fit right in.

Our values aren’t wall art — they describe how decisions actually get made here:

• Mission focused. The mission is our north star. We stay focused on what moves it forward and tune out the noise.
• Default to action. Move quickly and iterate. Progress over perfection. Learn by shipping.
• Solve with data. Instinct gets us asking the right…