Software Engineer (New Grad)

About Hyphenate

Most AI companies are building copilots.

Hyphenate is building the AI operating system for the CFO office. Our Audit-Ready AI Agents connect to a company's existing finance stack—ERPs, banks, billing, payroll, CRM, contracts, spreadsheets, email, and Slack—and automate the work behind the entire order-to-cash process, record-to-report process, treasury management, financial reporting, and audit readiness.

The goal isn't to help the finance team write better prompts.

The goal is for finance teams to review exceptions while AI does the rest.

What makes Hyphenate different is our Unified Finance Context—a financial understanding layer that captures transactions, policies, contracts, historical decisions, and accounting judgment. On top of it sit Audit-Ready AI Agents that can reason, explain their decisions, escalate uncertainty, and continuously improve.

We've raised $9M led by Foundation Capital, alongside BoldCap, Gaia Ventures, Aravind Srinivas (CEO of Perplexity), and finance leaders from Zuora, Ramp, Gusto, MongoDB, Zoom, and the Big Four.

What You'll Own

You won't get a starter ticket and a six-month ramp plan.

You'll join a pod of 2–3 engineers that owns a finance domain end-to-end—revenue, cash, close, reporting, payroll, fixed assets, tax, or controls—and you'll own real surface area inside it from your first month: context, prompts, tools, evals, guardrails, and the product around them. A senior engineer pairs with you closely at the start. The bar is that you're scoping and shipping your own work by the end of your first quarter.

No PM writes your specs. No architecture committee approves your ideas. You'll sit with the controllers, accountants, and CFOs who do the work today, understand it properly, then build the agent system that replaces it.

"New grad" is how you start here. It isn't a track, and it isn't a ceiling.

Problems Worth Your Brain

• How do you build AI agents that finance teams and auditors can trust? Verification, guardrails, observability, and evaluation systems for non-deterministic AI.

• How do you give an agent the right financial context? Context engineering over transactions, ledgers, contracts, policies, and historical decisions—without bloat, drift, or data leakage.

• How do you turn messy enterprise systems into a unified source of truth? Ingest, normalize, and reconcile data from ERPs, banks, payroll, billing platforms, CRMs, and email.

• How do you build agents that get better over time? Systems that explain their reasoning, escalate uncertainty instead of guessing, and improve measurably from human corrections.

• How do you orchestrate durable AI workflows in the real world? Long-running, replay-safe workflows across flaky, stateful enterprise systems.

• How do you safely write back to systems of record? Idempotent, audit-ready updates to ERPs and financial systems with full traceability.

A Few Strong Opinions

• The engineer who can't operate AI agents fluently is becoming obsolete, fast. Fluency with Claude Code, Cursor, and internal agent tooling is a second cortex. Come knowing how to use them; leave knowing how to build with them.

• "Backend vs. frontend" is dissolving. The agents do the typing. The constraint is product judgment, system design, and the ability to close the loop from problem to shipped feature. Our engineers ship full-stack when the work calls for it.

• The pod is the unit of leverage. Two engineers who can hold an entire module in their heads ship more than ten engineers who each own a slice. Specialization across pods, generalist within them.

• The accountant + engineer loop is the moat. Engineers who sit with controllers and build from what they see compound faster than the ones who don't. That starts on day one, not after you've "earned" customer access.

• We hire for slope as much as intercept—but the intercept still has to clear the bar. We care how fast you learn. We also care that you've already built things that were hard.

What Great Looks Like Here

• You learn fast and ship faster. You can pick up an unfamiliar codebase or an unfamiliar domain and be productive in days, not weeks. The ability to understand a finance workflow and translate it into software is the single most important skill on this team.

• You think like an owner. You're a current or future founder who scopes your own work, thinks from the customer's perspective, owns decisions, and drives outcomes without waiting for direction.

• You solve problems end to end. The team is split vertically. Every engineer makes decisions across the LLM pipeline, infrastructure, backend, and UX.

• You care about getting it right. A 100% solution beats an 80% one. When something breaks, you dig until you understand why—not just until it stops erroring.

• You operate AI at two levels. You use coding agents fluently to ship faster, and you have real craft in building the agents that are the product: prompt and context design, tool use, evals, and knowing when a model is the right tool and when plain code is.

• You communicate clearly. You ask good questions, share progress without being asked, and say "I'm stuck" early instead of late.

What You Should Have Done Before

• Finishing a bachelor's or master's degree, or graduated within the past year. Computer science, machine learning, or AI strongly preferred; adjacent quantitative degrees welcome if the work backs it up.

• At least one substantial software engineering or AI engineering internship. Early-stage and AI-native startups count for more here—if you've shipped to production at a company under 50 people, tell us about it.

• Hands-on experience building with AI agents. Internship, research, coursework, or a serious side project all count. We care that you've actually built one, watched it fail in an interesting way, and can explain what you changed.

• Real fluency with coding agents. Claude Code, Cursor, or equivalent. You have opinions about which to reach for and when, and you've shipped production code through them.

• Strong fundamentals in a modern language. Our stack is Python. If your depth is in C++, Java, Go, or Rust, strong fundamentals matter more—be ready to ramp fast.

• Something real you can walk us through in depth. A side project, a research prototype, an open-source contribution, a hackathon build. We'll go deep on it, so pick something you actually understand.

You'll Stand Out If

• You've done research in AI, ML, or systems—published or in progress at NeurIPS, ICML, ICLR, ACL, or similar. We weight real-world impact over citation count.

• You've TA'd or course-assisted a serious systems, ML, or compilers course. Teaching a hard thing well is a strong signal.

• You've shipped a side project that real people actually used—not just starred.

• You've contributed to open source that other people depend on.

• You've built evals, verification, or observability for something non-deterministic and watched it catch a bug before a human did.

• You've interned at a pre-seed through Series B startup and thrived in the ambiguity.

• You've done well in competitive programming or olympiads—ICPC, IMO, IPhO, Putnam, or similar.

• You're curious about fintech, accounting, or how money actually moves.

The Upside

• Exceptional teammates with high ownership and direct access to customers, CFOs, controllers, and founders.

• Competitive pay and meaningful early-stage equity.

• Full medical, dental, and vision coverage for employees and dependents, plus 401(k) match.

• Meals, a stocked NYC office, and the chance to help define an entirely new category: Audit-Ready AI for Finance.

The Details

• In-person in New York City. We build in the same room, on purpose.

• Visa status isn't a filter here. We support OPT and we go the distance on longer-term status for the engineers we hire. The paperwork is our problem, not yours.