Member of Technical Staff — Applied Machine Learning

Member of Technical Staff — Applied ML

New York, NY · On-site in Flatiron · Full-time

$175K–$300K base + highly competitive equity

The company

The company is building agents for accounting, starting with one of the most structured and high-stakes categories in business software.

Three years in, the product can complete a partnership tax workbook end to end.

The team is 80 people, fully in-person, and works around the corner from Madison Square Park in Flatiron. The company is backed by $100M in funding.

Accounting is a strong first domain for agents because the work is repetitive enough to automate, but consequential enough that reliability, traceability, and evaluation matter from day one.

The role

This is an applied ML ownership role for someone who wants to build the systems that let agents reason, plan, evaluate themselves, and improve in production.

You will own projects end to end: define the problem, design the system, run experiments, instrument the outcomes, and decide when the work is ready to ship.

The right person will be comfortable moving between model behavior, software architecture, and product constraints without waiting for handoffs.

This is not a role for someone who only wants to tune prompts or run isolated experiments. The scope includes the agent loop, evaluation, retrieval, guardrails, and the production systems around them.

The technical problem

The company is building agents that accountants can trust on real workflows.

That means the hard problems are not just model quality. They are context selection, tool use, fallback behavior, latency, cost, error analysis, regression detection, and the ability to measure improvement against real tasks.

The system has to work on messy accounting input, maintain coherence over multi-step tasks, and degrade gracefully when the model is uncertain.

The challenge is to build an agent platform that gets more capable over time without becoming opaque, brittle, or impossible to evaluate.

What you'll own

• Agent architecture: design and iterate the systems that let agents plan, call tools, recover from failures, and complete accounting workflows reliably.

• Evaluation infrastructure: build offline and online evaluation pipelines that can run hundreds of experiments, define gold tasks and metrics, and make model changes comparable.

• Error analysis and monitoring: instrument the stack to detect regressions, surface failure modes, and turn production data into concrete product and architecture decisions.

• Retrieval and context systems: build prompt hierarchies, retrieval pipelines, and context packaging that give models the right information at the right time.

• Document understanding: parse messy financial and accounting documents into structured representations that agents can reason over.

• Guardrails and validation: design boundaries that keep behavior deterministic where it needs to be and safe where mistakes are expensive.

• Model routing and optimization: choose when to use which model, and optimize for latency, cost, and accuracy in real production conditions.

• Project ownership: scope work clearly, write concise specs, ship end to end, and communicate what changed, what improved, and what still does not work.

Who this is for

You are likely a strong fit if you have:

• Built and shipped production ML systems, not just prototypes or notebooks.

• Owned an evaluation stack or experimentation framework where model quality was measured, not argued about.

• Worked on agents, search, retrieval, ranking, information extraction, or other systems where the model sits inside a larger product loop.

• Strong judgment about context management, tool use, prompt structure, and failure recovery.

• Comfort reasoning about latency, cost, accuracy, and reliability together.

• Experience turning messy real-world data into structured inputs that models can use.

• The ability to read ambiguous product requirements and turn them into an executable technical plan.

• The habit of using experiments and error analysis to drive decisions.

• Comfort working in a small, high-ownership team where the people building the system also own the outcome.

Tech stack

• Primary languages: Python

• Data: Postgres

This role lives primarily in Python and Postgres. The leverage comes from the systems you build on top of them, not from adding more infrastructure for its own sake.

Why now

The company has already crossed the line from experimentation to real production use.

The next problem is not whether agents can do useful work at all. It is whether the system can become measurably better every month while staying reliable on real accounting tasks.

That requires stronger evaluation, better context systems, better failure handling, and a tighter loop between production data and product decisions.

The work done now will shape how the system behaves as usage expands and as the accounting surface area becomes more complex.

This role is not for you if

• You want a narrowly defined IC role with clean boundaries around ownership.

• You prefer research work that does not need to survive production constraints.

• You are not comfortable making tradeoffs when quality, latency, and cost pull in different directions.

• You want to work remotely.

• You do not want to own the messy parts of the system after the demo works.

Compensation and logistics

• Base salary: $175K–$300K

• Equity: highly competitive

• Location: New York, NY

• Work model: on-site, 5 days per week in Flatiron

• Visa sponsorship: available

• Employment: full-time

Interview process

Typical process:

• Initial screen

• Meet the team

• Technical interview with a coding component

• Onsite plus references

About Aurora

Aurora helps exceptional engineers find the right role at some of the most ambitious startups worldwide.

We work with teams that value high ownership, strong technical standards, and clear impact.