Build the marketplace that pays experts what their judgment is worth.

Marketplace owns everything an expert touches between finding a project and being matched to one: the Explore board, search and filtering, the shared application checklist, the screening flow, and the state machine that moves an application from submitted to matched to active. Roughly 3,400 applications a week pass through it, and about a third stall somewhere in the middle — usually at an identity check or a skills assessment that timed out.

The immediate work is unglamorous and high-leverage. The application checklist is currently one component holding eleven pieces of state, and every new lab program wants a variant of it. We want it decomposed into a declarative step definition so that adding a program-specific step — a language screen, an NDA, a licensure upload — is a data change rather than a release.

You would be the fifth engineer on the team and the second senior one. There is no separate front-end or back-end here: you will write the React, the Postgres queries behind it, and the background jobs that reconcile the two when they disagree.

• ✓Own the application state machine end to end — schema, transitions, background reconciliation, and the UI that renders it
• ✓Rebuild the application checklist as a declarative step definition so new lab programs ship without a code change
• ✓Cut the stall rate on in-progress applications, currently around 34%, by instrumenting each step and fixing the three worst drop-offs
• ✓Design and run the search and filtering layer on the Explore board as the catalogue grows past a few hundred concurrent programs
• ✓Write the migrations, run them yourself, and keep them reversible — we do not have a DBA
• ✓Carry a one-week primary on-call rotation roughly every seven weeks, covering marketplace services only
• ✓Review other engineers' pull requests with actual comments, and take the same in return
• ✓Write a one-page design memo before any project expected to take more than two weeks

• ✓Five or more years building and operating production web applications, with real ownership of at least one system that other teams depended on
• ✓Strong TypeScript and React — you are comfortable reasoning about rendering, state colocation and why a component re-rendered
• ✓Practical relational database skills: you can read a query plan, add the right index, and explain why a migration is or is not safe under load
• ✓Experience with long-running, multi-step user flows where partial state and retries are the norm rather than the exception
• ✓You have debugged a production incident to root cause and written the follow-up yourself
• ✓Clear written English — most decisions here are made in a document, not a meeting

• ✓Next.js App Router in production, including the parts that are still awkward
• ✓Experience with marketplaces, two-sided supply, or anything where the hard problem was matching rather than throughput
• ✓Background jobs at scale — queues, idempotency, exactly-once illusions and their failure modes
• ✓You have worked somewhere between 20 and 150 people and can say what broke as it grew

Every Friday, Cortext runs a payout batch across 130 countries through Stripe and Wise. The batch is the single most consequential piece of software in the company: an expert who is not paid on time does not come back, and a payout run that half-lands is materially worse than one that fails outright and retries cleanly.

The platform team owns that batch, the double-entry ledger behind it, the tax-form collection that gates it (W-9 for US contributors, W-8BEN for everyone else), and the identity verification that decides whether a payout account may exist at all. Today the ledger and the payout executor share a database and too many assumptions. We want them separated, with the ledger as the system of record and the executor as a dumb, aggressively idempotent consumer.

This is a staff role because the hard part is not writing the code — it is sequencing a migration of live money movement without a maintenance window, and being the person who can say with evidence that Friday will be fine.

• ✓Own the payout pipeline: ledger, batch executor, rail adapters for Stripe and Wise, retry and reversal handling
• ✓Separate the ledger from the executor and migrate live balances without pausing a payout week
• ✓Drive failed-payout rate down from the current 1.9% of transfers, most of which are stale or mistyped beneficiary details
• ✓Build the reconciliation that proves, every Monday, that the ledger and both rails agree to the cent — and alert when they do not
• ✓Own the tax-document gate: collection, validation, expiry and the blocking rules that keep an unverified account from being paid
• ✓Set the engineering bar for the platform group: design review, incident review, and what we will not build
• ✓Partner with Finance on 1099-NEC generation and with Trust & Safety on account-takeover response
• ✓Carry primary on-call for money movement, including the Friday batch window

• ✓Eight or more years in backend engineering, with at least three on payments, ledgers, billing or another domain where being wrong costs money
• ✓You have built or substantially rewritten a double-entry ledger and can explain why the naive schema fails
• ✓Deep experience with idempotency, exactly-once semantics and the compensating actions that replace them in practice
• ✓You have migrated a production system carrying real financial state, and can describe the rollback plan you actually needed
• ✓Comfort with third-party payment rails and their asymmetries — settlement timing, partial failures, and webhooks that arrive twice or never
• ✓Willingness to be on call for the payout window, and the judgment to know when to stop a batch
• ✓Written communication strong enough that Finance and Trust & Safety can act on your documents without a meeting

• ✓Cross-border contractor payouts, or anything involving KYC and sanctions screening at volume
• ✓Experience with Wise or Stripe Connect specifically, including their less-documented edges
• ✓Familiarity with US contractor tax reporting (1099-NEC, W-8BEN, backup withholding)
• ✓You have been the person who wrote the postmortem for a payment incident that reached customers

A lab partner sends us a task specification; we turn it into batches an expert can actually complete, route them, collect the submissions, run quality checks, and deliver a dataset back under a contractual schema. That round trip is the product, and it currently runs on a mix of Airflow DAGs, a Postgres staging schema and three scripts nobody wants to own.

Volume is the forcing function: submissions crossed 210,000 in the last quarter across 40+ lab partners, and every partner wants a slightly different delivery format, redaction rule and acceptance threshold. The team's job is to make partner-specific delivery a configuration rather than a fork.

You will spend as much time with Research Operations as with engineers. The people who know why a batch failed are usually programme managers, not services.

• ✓Own the ingestion-to-delivery pipeline for lab task data: batching, routing, submission capture, QA sampling, packaging and hand-off
• ✓Replace per-partner delivery scripts with a declared schema plus validation, so a new partner is onboarded in days rather than weeks
• ✓Build the QA sampling layer — gold-standard task injection at roughly 4% of volume, inter-rater agreement scoring, and the alerts when agreement drops
• ✓Instrument batch health so Research Operations can see, without asking, which batches are behind and why
• ✓Enforce redaction and retention rules on submission data, including partner-specific deletion windows
• ✓Backfill and reprocess historical batches when a guideline changes mid-programme, without corrupting delivered datasets
• ✓Keep pipeline cost visible and roughly flat as volume grows

• ✓Four or more years building production data pipelines that other people depended on being correct
• ✓Strong Python and SQL; you write pipelines that are testable rather than pipelines that happen to have worked so far
• ✓Experience with a workflow orchestrator (Airflow, Dagster, Prefect or equivalent) including its operational failure modes
• ✓Data modelling for evolving schemas — you have handled a breaking upstream change without a full rewrite
• ✓Comfort owning data quality as a first-class product concern, not a downstream complaint
• ✓You can work directly with non-technical operators and translate their descriptions into checks

• ✓Human annotation, labelling or evaluation data specifically — inter-rater agreement, adjudication, rubric drift
• ✓dbt, or another opinionated transformation layer, in a team of more than two
• ✓Experience with data deletion and retention obligations under a customer contract

Cortext is paid for expert judgment, so the question we cannot avoid is: how good is this submission, and how do we know? There is no ground truth for whether a written rationale is a good rationale. We approximate it with seeded gold-standard tasks, inter-rater agreement, adjudication by senior reviewers, and a set of learned signals that flag work for human review.

This role owns those learned signals. Concretely: scoring models that rank submissions for QA sampling, drift detection when a guideline changes mid-programme, and detection of submissions that were themselves generated by a language model — which, on an AI-training marketplace, is the failure mode that most directly damages the product we sell.

The constraint that shapes everything: a false positive costs a real person their income. Every flag this team produces routes to a human reviewer with the evidence attached, and we hold ourselves to a precision target rather than a recall one.

• ✓Build and maintain the submission scoring models that decide which of roughly 8% of sampled work gets human QA
• ✓Own LLM-generated-submission detection, including the evaluation set, the precision target and the review path for every flag
• ✓Detect calibration drift across reviewers and programmes, and surface it before a delivery misses its acceptance threshold
• ✓Design labelling and adjudication processes with Research Operations to generate the training data you need
• ✓Ship models as services the marketplace can call synchronously, with latency budgets and sane fallbacks
• ✓Run offline and online evaluation honestly, including publishing the cases where a model made things worse
• ✓Write the appeal path documentation with Trust & Safety, so an expert can contest a flag with a human

• ✓Five or more years in applied ML, with models you shipped and then maintained through their decay
• ✓Solid grounding in evaluation: precision/recall tradeoffs, threshold selection under class imbalance, and why offline metrics mislead
• ✓Practical NLP experience, including fine-tuning or adapting language models for classification or scoring
• ✓You have built a system where a false positive harmed a person, and can describe the guardrails you added
• ✓Strong Python engineering — your models run in production without a separate team to rewrite them
• ✓Comfort designing annotation processes rather than assuming labelled data appears

• ✓Experience with RLHF, preference data, or model evaluation pipelines at a lab or data vendor
• ✓Anomaly and fraud detection where adversaries adapt to your detector
• ✓Published or internal work on inter-rater agreement and rubric calibration

A lab programme is a contract with a schema, an acceptance threshold, a delivery calendar and a cohort of experts who have to be recruited, briefed, calibrated and kept. Delivery managers own that from kickoff to final hand-off. You would run three to five concurrent programmes, typically 40 to 300 experts each, worth $200,000 to $2M in delivered work over six to twelve weeks.

The recurring failure is not effort, it is calibration. A lab changes its rubric in week three; agreement between reviewers collapses; the delivery misses its acceptance threshold; and nobody notices until the weekly report. The job is to notice in day two, re-brief the cohort, and renegotiate the schedule with the lab before it becomes a credit note.

This is an operating role, not a coordination role. You will be reading submissions yourself, sitting in on adjudication calls, and telling a partner that their specification is ambiguous.

• ✓Own three to five concurrent lab programmes end to end: kickoff, cohort build, calibration, delivery, retrospective
• ✓Translate a partner's task specification into briefs, rubrics and worked examples an expert can act on without a call
• ✓Run calibration: seed gold-standard tasks, monitor inter-rater agreement weekly, and re-brief cohorts before agreement drops below the acceptance threshold
• ✓Hold the delivery schedule, including the uncomfortable conversation when a partner's mid-programme change makes it unachievable
• ✓Work with Expert Supply on cohort composition, attrition and backfill — cohorts typically lose 12–18% over a ten-week programme
• ✓Escalate quality disputes and pay disputes to the right owner within one business day, and close the loop with the expert
• ✓Run a written retrospective on every programme, with the numbers, and feed the fixes into the next one
• ✓Keep the programme margin honest: flag scope creep in writing before absorbing it

• ✓Four or more years running operational programmes with external stakeholders and a real delivery deadline
• ✓Comfort with data: you can build the agreement report yourself in SQL or a spreadsheet rather than filing a ticket for it
• ✓Demonstrated ability to write instructions that non-specialists follow correctly the first time
• ✓Experience managing a distributed contributor or contractor population, including attrition and quality variance
• ✓Directness with clients — you have told a paying customer that their request is not possible, and kept the account
• ✓Availability that overlaps the US East Coast working day, with occasional flexibility for London and Bengaluru handoffs

• ✓Prior work in data labelling, annotation, market research fieldwork, or clinical trial operations
• ✓Exposure to AI or ML programmes, including familiarity with evaluation rubrics
• ✓A second language relevant to our expert base — Spanish, Portuguese, French, Hindi or Swahili

The Nairobi office runs the Africa expert network: onboarding, verification support, community, quality coaching and the day-to-day of keeping several thousand contributors working well across the continent. It is the fastest-growing part of the expert base and the one where the standard playbook fits worst — payment rails behave differently, connectivity is uneven, and a mobile-first contributor experience is the default rather than an accommodation.

You would lead the team and own the outcomes: time from application to first paid task, first-90-day retention, and quality scores relative to the global cohort. Today the median African contributor waits longer than the global median at the identity-verification step, and closing that gap without weakening the check is the first order of business.

This role has real authority to change process, and a corresponding obligation to argue for it with evidence at the global operations review.

• ✓Lead and grow the Nairobi operations team, including hiring, performance and career development for six to ten people
• ✓Own onboarding throughput for the Africa network: application to first paid task, currently a median of 14 days against a global median of 9
• ✓Run quality coaching for contributors flagged by QA, with a documented path back to good standing rather than silent deactivation
• ✓Work with the Platform team on payout reliability across African corridors, and represent the corridor-specific failures that get lost in the global average
• ✓Build community: local expert meetups, a moderated forum, and a first-response channel with a same-business-day target
• ✓Partner with Trust & Safety on identity fraud patterns specific to the region, without letting suspicion default onto honest contributors
• ✓Report monthly on network health with numbers the rest of the company can act on

• ✓Six or more years in operations, community or customer teams, including at least two managing people directly
• ✓You have run a distributed contributor, agent or gig population of at least 1,000 people
• ✓Track record of process design that survived contact with volume — documented, measured and revised
• ✓Fluent written English and the ability to hold your position in a room where the loudest voices are in San Francisco
• ✓Comfort with data: you build your own reports and defend the definitions in them
• ✓Existing right to work in Kenya

• ✓Swahili, French, Amharic, Yoruba or Hausa in addition to English
• ✓Experience with mobile money rails and their reconciliation quirks
• ✓Prior work with an international company where the head office was in a different timezone and a different mood

Trust & Safety decides who is allowed to earn on Cortext. The caseload splits roughly into three: identity fraud (one person operating several accounts, or an account operated by someone other than the verified holder), work fraud (submissions produced by a language model and presented as expert judgment), and account takeover.

Every case ends in a decision that affects someone's income, so the standard is evidentiary rather than intuitive: a written case file, the specific signals relied on, the contributor's response where policy allows one, and a decision an appeals reviewer can audit six months later. We publish our reversal rate internally and treat a high one as a signal about our own thresholds.

This is deliberately not a queue-clearing role. Investigators are expected to spot the pattern behind a cluster of cases and write the policy or detection change that removes it, and roughly a fifth of your time is reserved for exactly that.

• ✓Investigate escalated fraud, identity and account-integrity cases to a written, auditable standard
• ✓Decide enforcement — warning, restriction, suspension or permanent removal — and communicate it in plain language the contributor can act on
• ✓Handle appeals on cases you did not decide, and reverse them when the evidence does not hold
• ✓Identify clusters and repeat patterns, and convert them into detection rules with the Quality Signals team
• ✓Maintain the case-file standard and keep decision quality consistent across the team through weekly calibration sessions
• ✓Contribute to policy: propose the wording change when a rule proves unenforceable or unfair in practice
• ✓Meet the same-day first-response target on safety reports raised through the Trust & Safety channel

• ✓Three or more years in trust and safety, fraud investigation, financial crime, content moderation policy or a comparable investigative role
• ✓Demonstrated ability to write a case file that a stranger could audit without asking you questions
• ✓Sound judgment under ambiguity, and the discipline to record uncertainty rather than resolve it with confidence you do not have
• ✓Comfort querying data — you can pull the account graph yourself rather than requesting it
• ✓Resilience for the material: this role involves reviewing deception, occasional abusive communication, and, on red-team programmes, distressing model output
• ✓Working hours overlapping European business hours, and existing right to work in an EMEA country where we can employ

• ✓Experience with KYC, sanctions screening or document verification vendors
• ✓Familiarity with LLM-generated text detection and its limits
• ✓A second European language for contributor correspondence
• ✓Prior work on an appeals or adjudication function

Expert Supply is recruiting, except the candidates are contributors, the roles are lab programmes, and the cycle time is days. When a partner opens a programme needing 60 licensed pharmacists across three timezones inside two weeks, this team either finds them or the programme does not run.

The APAC role owns supply across India, Southeast Asia, Japan, Korea and Australia — the fastest-growing region by applications and the one where credential verification is hardest, because degree and licence registries differ by country and several are not queryable at all.

The work is half sourcing channel management (universities, professional bodies, alumni networks, targeted paid acquisition) and half conversion: the drop between application and first paid task is where most of the value is lost, and it is measurable to the day.

• ✓Own APAC supply targets by speciality: medicine, law, engineering, quantitative finance and languages
• ✓Build sourcing channels beyond paid acquisition — professional bodies, university networks, specialist communities — and report cost per activated expert per channel
• ✓Run rapid cohort builds against programme deadlines, typically 40 to 200 verified experts in ten working days
• ✓Own credential verification workflow for APAC qualifications, including the countries where no central registry exists
• ✓Reduce application-to-first-task drop-off, currently 41% in the region, with interventions you can prove worked
• ✓Manage two supply associates, and grow the team as regional demand grows
• ✓Feed realistic supply forecasts into Go-to-market so we do not sell a cohort we cannot staff

• ✓Five or more years in recruiting, talent operations, marketplace supply or high-volume sourcing
• ✓You have hit a hard numeric hiring or activation target repeatedly, and can describe the funnel that got you there
• ✓Experience owning a funnel metric end to end, with your own reporting
• ✓Credential or background verification experience in at least one regulated profession
• ✓Excellent written English and the judgment to write outreach that does not read as spam
• ✓Existing right to work in India

• ✓Experience recruiting clinicians, lawyers or other licensed professionals
• ✓Prior work at a marketplace, staffing platform or gig platform
• ✓Working knowledge of a second APAC language
• ✓Familiarity with paid acquisition mechanics and their measurement

Cortext sells to a small, technically sophisticated set of buyers: research and data leads at frontier AI labs and at the largest applied-AI teams inside enterprises. There are perhaps 120 accounts on earth worth calling, we already work with 40+, and the buyer can tell within ten minutes whether you understand evaluation.

Deals are $250,000 to $4M in annual committed spend, run three to six months, and are won on delivered quality rather than on a deck. Expect a paid pilot in almost every cycle: a small cohort, a real specification and an acceptance threshold you will be measured against.

The quota for this role is $4.5M in new committed spend in year one, with a ramp of two quarters. We publish the quota, the accelerators and the commission plan before you sign, and we do not change them mid-year.

• ✓Own a named list of 25 to 40 frontier lab and enterprise AI accounts end to end, from first contact to signed agreement
• ✓Run technical discovery deep enough to write the pilot specification yourself with Research Operations
• ✓Design and close paid pilots that convert — the current pilot-to-contract rate is 46%, and it is the number this role moves
• ✓Negotiate commercial terms with procurement and legal, including data handling, retention and indemnity provisions
• ✓Forecast honestly, in writing, weekly — an inflated forecast costs Expert Supply a cohort they cannot unbuild
• ✓Bring competitive and product feedback back with enough specificity to be actionable
• ✓Support renewal and expansion with the delivery team through the first year of every account you close

• ✓Six or more years closing enterprise deals above $250,000 annual value, with quota attainment you can evidence
• ✓Genuine technical fluency in AI and ML — you can hold a conversation about evaluation, fine-tuning and data quality without a solutions engineer
• ✓Experience selling a service or data product where delivery quality, not features, was the deciding factor
• ✓You have negotiated with enterprise procurement and legal, and know which terms are worth escalating
• ✓Disciplined written forecasting and pipeline hygiene
• ✓Willingness to travel roughly 25% of the time, mostly domestic US

• ✓An existing network at AI labs or applied research teams
• ✓Prior experience at a data, annotation or evaluation vendor
• ✓Comfort with usage-based or committed-spend commercial models

Partnerships is the other side of supply. The fastest route to 200 verified radiologists is not paid acquisition — it is an agreement with a professional body, a university department or a specialist association that already has them and is willing to say we are legitimate.

This role owns those agreements across Europe, the Middle East and Africa: institutional partnerships that bring credentialled experts into the network, plus the co-marketing and revenue-share terms that make them worth signing for both sides. The London office already anchors the EMEA expert network, so you would be building on an existing base rather than from zero.

It also owns the harder conversation: what we will not agree to. Exclusive supply arrangements, data-sharing terms that conflict with our privacy commitments, and referral structures that read as payment for credentials are all off the table, and part of the job is saying so early.

• ✓Source, negotiate and close institutional partnerships across EMEA — professional bodies, universities, specialist associations and alumni networks
• ✓Own a target of 12 signed partnerships in the first year, measured on activated experts rather than logos
• ✓Structure commercial terms including referral fees, co-marketing and data handling, with Legal and Finance
• ✓Build the partner onboarding path with Expert Supply so a signed agreement produces verified contributors within 30 days
• ✓Represent Cortext at professional and academic conferences, and run the follow-up that makes the trip worth it
• ✓Maintain relationships after signature — the second-year renewal is where most partnerships quietly fail
• ✓Say no clearly and early to terms that conflict with our privacy, exclusivity or fairness commitments

• ✓Six or more years in partnerships, business development or institutional sales, with agreements you owned from sourcing to signature
• ✓Experience negotiating with institutions rather than only companies — universities, regulators, professional bodies and their committee timelines
• ✓Ability to translate a commercial structure into terms Legal can draft without a translation layer
• ✓Strong understanding of EMEA data protection expectations, particularly around candidate and professional data
• ✓Excellent written English; a second European language is a real advantage
• ✓Existing right to work in the United Kingdom, and willingness to travel roughly 30% of the time across EMEA

• ✓An existing network among European professional bodies or university faculties
• ✓Experience in regulated professional markets — medicine, law, accountancy or engineering
• ✓Prior work at a marketplace or credentialling platform

Cortext paid $18.4M to experts in 2025 across 130 countries, in weekly batches, to people who are contractors rather than employees. That produces a month-end close with an unusual shape: high transaction volume, many currencies, two payment rails, and a tax-reporting obligation that lands as a single hard deadline every January.

This role owns contractor payments accounting: accruals for work completed but not yet paid, reconciliation between the internal ledger and both rails, foreign exchange treatment, and the 1099-NEC cycle. It also owns the boring, load-bearing work of keeping W-9 and W-8BEN records complete enough that January is a process rather than an emergency.

Close is currently eight business days. The first objective of this role is six, without loosening a single control.

• ✓Own the contractor payments cycle in the close: accruals, cut-off, reconciliation to Stripe and Wise, and unresolved-item follow-up
• ✓Reconcile the internal ledger to both payment rails weekly and investigate every variance, not only material ones
• ✓Own the annual 1099-NEC cycle end to end, including TIN matching, corrections and contributor queries
• ✓Maintain W-9 and W-8BEN completeness, and work with Platform on the gating rules that keep an unverified payee unpaid
• ✓Handle multi-currency accounting and FX revaluation for non-USD payouts
• ✓Cut close from eight business days to six by automating reconciliation rather than by skipping steps
• ✓Support the first external audit, including documenting controls that currently exist only in practice
• ✓Answer contributor payment queries escalated by Support with an accurate answer within one business day

• ✓Five or more years in accounting, including US GAAP month-end close ownership
• ✓Direct experience with contractor or marketplace payouts at volume, or with a high-transaction payments environment
• ✓Working knowledge of US contractor tax reporting: 1099-NEC, W-9, W-8BEN, backup withholding and TIN matching
• ✓Strong reconciliation discipline and genuine comfort with large datasets — advanced spreadsheet skills, and SQL is a real plus
• ✓Experience with multi-currency transactions and FX revaluation
• ✓Clear written communication with people who are not accountants, including contributors asking why a payment differs from their expectation

• ✓CPA, or an equivalent qualification in progress
• ✓Experience preparing for a first external audit at a venture-backed company
• ✓Familiarity with Stripe Connect or Wise reporting exports
• ✓Exposure to non-US contractor reporting obligations

You would be the second designer, working on everything an expert sees: the Explore board, the application checklist, the screening flow, the task workspace where the actual work happens, and the earnings and payout screens where trust is either built or lost.

The interesting constraint is that our users are experts at something other than software. A consultant radiologist reading a rubric at 11pm on a phone is not going to explore the interface; the design has to be unambiguous on first read, in a second language, on a mid-range Android device, on an uneven connection.

The second constraint is money. Screens that explain what someone earned, why a submission was rejected, or when a payout will land are the highest-stakes surfaces in the product. They are also, today, the least designed. Fixing that is the first project.

• ✓Own design for the expert-facing product end to end: research, interaction, visual, and the specifics engineering needs to build it
• ✓Redesign the earnings, submission-feedback and payout surfaces so an expert can answer what was paid, what was rejected and why, without contacting support
• ✓Run lightweight research with real contributors — six to eight sessions per project, including non-native English speakers and mobile-only users
• ✓Design for a mobile-first, low-bandwidth reality, and pressure-test on a mid-range Android device rather than a simulator
• ✓Extend the design system with the second designer, and keep it honest about what is actually implemented
• ✓Bring accessibility into the default: contrast, focus order, screen-reader labelling and keyboard paths, tested rather than assumed
• ✓Write the copy in your flows first, then refine it with the team — the words are the interface here

• ✓Four or more years designing production software, with shipped work you can walk through in detail
• ✓A portfolio showing complex, multi-step flows rather than marketing pages — we care about the third and fourth screen
• ✓Genuine interaction design skill, plus enough visual craft to make it feel finished
• ✓You have run your own user research and changed your design because of it
• ✓Ability to write clear product copy, including error and rejection messaging
• ✓Comfort working closely with engineers, including reviewing implementation and pushing back on drift

• ✓Experience designing for users outside your own language and market
• ✓Accessibility work you can evidence, ideally to a WCAG standard
• ✓Design systems work in a small team where you had to keep it small
• ✓You can prototype in code well enough to test an interaction