Product Engineer

About Refine

A spreadsheet error in one of the most influential economics papers of the decade helped justify austerity across Europe and the US. Millions lived with the consequences. The error was only found years later, by sheer luck, long after the damage was done.

Flawed analysis shapes policy, investment and lending every day, costs billions every year, and mostly goes unnoticed. AI makes it worse: rigorous-looking work is now cheap to produce.

Refine exists to catch these errors before anyone acts on them. We're building the infrastructure that verifies high-stakes work; think CI/CD for everything beyond code. It reviews a document like a team of experts, checking the maths, data, logic and argument, in minutes.

• The best AI review system. Refine 5 beats frontier models such as GPT-6 Astra in over 90% of blind head-to-heads across ten fields, from economics to biology. It combines the top frontier models at hundreds of times normal compute, so no one model's blind spots get through. We tune it on documents seeded with the errors frontier models miss.

• The de facto AI reviewer for economics. Econometrica and the AER run every paper they publish through Refine. It's used at 24 central banks and by researchers at Oxford, MIT, Stanford, Yale and Cambridge; Cambridge, UCLA and NYU provide it to theirs.

• Beyond academia. 29,000+ reviews sold at 1,000+ institutions, including litigation consultancies, hedge funds and companies bidding on public tenders.

• Enterprise-grade. ISO 27001 certified, SOC 2 Type 1 (Type 2 in progress). On the web and inside ChatGPT, Claude and Codex.

The role

You'll build Refine with a swarm of AI agents: they write most of the code, and you do the thinking. You own work end to end, from the customer who asked for it, to the code running in production, to the numbers that say whether it worked.

The work covers the whole product:

• The core AI pipeline. The review engine itself: new checks, how frontier models are combined, and what happens when one of them fails.

• Organizations and B2B. Self-serve organizations, team and department plans, institutional checkout, and what large clients need next, such as single sign-on.

• Pricing. New plans, packs, discounts and trials, end to end through payments.

• Integrations. Refine already runs inside ChatGPT, Claude and Codex, and there's more to connect.

You'll make the everyday technical calls yourself, so the founders can spend their time on customers and growth. You'll work with Nacho García, who leads engineering, and with the founders, Yann Calvo Lopez (CEO) and Ben Golub (Chief Scientist, formerly a tenured economics professor at Harvard). Our advisors include OpenAI co-founder John Schulman.

How we build

This is the part we're proudest of. The codebase is built for people and agents to work on together, and it holds up.

• Cloud native, everything in code. We run on managed cloud services, and every environment is infrastructure as code in Pulumi: production, an EU stack, and a throwaway stack for any branch.

• Domain-driven design and hexagonal architecture. Business rules live in a domain layer with no infrastructure in it. Aggregates raise their own domain events, and long workflows run as durable sagas. Architecture tests enforce the boundaries, so nobody, human or agent, quietly breaks them.

• A real testing pyramid. Unit tests at the base. An acceptance test for every use case, against a real database. Whole review pipelines run end to end with scripted models. Browser end-to-end and accessibility tests at the top. Every pull request gets a security scan and an AI review before a person looks at it.

• Measured before it ships. We benchmark changes to models, prompts and checks against the current pipeline and against one call to the strongest frontier model, then roll them out behind feature flags.

• Built for LLMs that fail. A review makes many calls to frontier models from several providers. Workflows survive crashes and pick up where they stopped. Every call and stage has its own time budget. Stages degrade gracefully, with fallbacks. If a review can't be completed, the customer gets their credits back automatically.

• Private by default. Model traffic only goes to providers that don't store our data.

• Enterprise-grade, because our clients are. ISO 27001 certified and SOC 2 Type 1, with Type 2 in progress and EU data residency on the way.

Our data, and AI on top of it

Every domain event goes to PostHog, sent only once the change it describes is committed, and so does every model call, with its stage, model, cost and latency. Next to them sit 18 warehouse sources:

• payments and subscriptions;

• sign-ins and organizations, and our app database;

• email and CRM;

• three ad platforms and search;

• GitHub;

• logs;

• the bills from our model providers and our cloud.

Ad click to revenue is one query. So is what a review cost us against what the customer paid. So is what a change behind a flag did, per user or per organization.

• One definition per number. What a number means is defined once, in a metric catalog a person approves: revenue, sign-up to paid, second-review rate, cost per new customer by ad channel, and more. AI agents use those definitions, so their answers agree.

• Self-serve, with AI. Anyone at Refine asks in plain language, such as "is the model swap we shipped on Monday working?" or "what do repeat users do differently?", and gets the answer with the query behind it.

• AI across the company: development, data analysis, production, marketing, hiring and support. Making that setup better is part of the job.

What you'll do

• Run a swarm of agents. Design the change, split it up, let agents build it in parallel, and review what comes back as if you'd written it yourself.

• Own features end to end. Talk to the users who asked, build, ship behind a flag, measure, iterate.

• Work across the stack: the review pipeline, the API, the web app and the infrastructure.

• Keep the bar high. Tests at the right level, clean boundaries, and nothing merged that you can't explain.

• Make the machine faster. Write the skills, rules and tools that make every engineer and agent here better.

Who you are

• AI-native, for real. You build with coding agents (Claude Code, Codex or similar) every day, several at a time, and can show us something real you shipped that way.

• Strong on architecture and design patterns. You know domain-driven design, hexagonal architecture (ports and adapters), event-driven design and the classic patterns, and when not to use them. You can tell an agent where new code belongs, and you notice when it puts code in the wrong layer.

• An engineer with judgement. You know what good tests look like at every level of the pyramid, and why design and tests matter more when agents write most of the code.

• Product-minded. You care whether what you built gets used, and you'll talk to users to find out.

• An owner. From idea to production to measured result, without waiting to be told.

• Low ego, clear writer. We're direct with each other, and our work is written down: designs, plans, pull requests. Fluent in English.

Language, title and years of experience don't matter. Our stack today is Python and TypeScript on Azure, with Pulumi for the infrastructure. If you've never written them, that's fine.

Nice to have: sagas, durable workflows or event-driven systems in production; infrastructure as code (Pulumi or similar); LLM pipelines or evals; payments and B2B SaaS (Stripe, single sign-on); PostHog or similar; Spanish.

Probably not for you if

• You want a spec handed to you.

• You'd rather write every line by hand.

• You want a calm, predictable roadmap. Things move fast here and sometimes get messy.

How we work

• AI-native throughout. One person with the right setup does what used to take a team.

• Small team, strong engineering. Everyone executes, founders included, and engineering ships fast.

• Autonomy. You choose what to work on, by impact.

• Speed. We build at the speed of AI.

• Remote-first. We're spread across Europe and the US. We work async by default, with regular team syncs, frequent customer calls and the occasional in-person offsite.

• Respect for users. They're sceptical experts; we win them with substance, not hype.

• Meritocracy. Talent, hard work and results are rewarded.

Compensation: competitive.

How we hire

1. Written answers to the questions below. Using AI to polish them is fine; the thinking has to be yours.

2. An exercise.

3. Conversations with the founders.

4. A paid trial: one or two days working with us on real work.

How to apply

Look around refine.ink, then send your CV, LinkedIn or GitHub with short answers to:

1. Show us something you built mostly with AI agents. What did you decide, and what did the agents do?

2. How do you make sure code written by an agent is correct and well designed?

3. What's one thing you'd change about Refine? Why, and how would you know it worked?

How you think matters more than polish.