Great minds. Deep human judgment.
Bring human judgment, taste, and the subtleties no manual captures to the next generation of intelligence.
Controllers
Who know when a close is really closed.
Traders
Who know a lucky day from a good strategy.
Underwriters
Who know which risks should never be written.
Engineers
Who know working code from code that merely passes.
Researchers
Who know which result to trust.
The network
reachable, verifiable, high-quality professionals
Humanity’s hard-won experience, within reach.
Every field. Every stage of a career. Alumni of the university clubs, societies and institutions we partner with — real people, with credentials you can check.
- Universities in our network
- 21
- Experts on the waitlist
- 7,000+
- Every level of experience
- Every field
Connecting
Stanford University
Data and products supported by people from
University logos show a selection of the 21 universities in our network. Logos identify where members of our network studied or work, and the universities whose student societies partner with us. They do not imply endorsement by these institutions or firms.
Waitlist
From zero to seven thousand.
Practitioners who asked to help define what correct means in their field.
Join them.
Leave your email and field. We invite experts in waves, as worlds in each field open.
Already know the role you want? Apply for it directly.
Who we are recruiting
Practitioners who have done the work
People who have closed the books, cleared the trades and signed the reports.
Shape what machines count as correct
We write the rule once with you, then it grades a million runs.
38 roles
T1Markets & executionTop rates $150–300 / hr
01Portfolio ManagerHedge fund or prop PM with a verifiable record
You will decide whether a strategy’s edge is real. Sizing, drawdown, turnover and costs: the difference between a lucky quarter and a process that repeats.
What we are looking for
- Track record — Managed capital at a hedge fund, prop firm or bank
- Risk discipline — You size for the drawdown, not the upside
- Scepticism — You distrust a backtest until it survives costs and a withheld period
- Precision — You can turn a judgment into a rule code can check
Work you might do
- Defining when a result counts as skill rather than luck
- Writing reference strategies that agents are measured against
- Finding strategies that pass the grader without real edge, and telling us
02Market MakerProp desk, exchange, or bank desk
You will encode what separates a quote that filled from a quote that never touched the book. Walking the book, capacity limits, slippage, adverse selection — the constraints that decide whether a signal ever became a trade.
What we are looking for
- Desk experience — Current or former market maker at a proprietary trading firm, exchange, or bank desk
- Microstructure fluency — You can explain why an order expired rather than filled, and what that cost
- Precision — Comfort specifying a rule tightly enough that code can check it
- Integrity — Willingness to tell us when our verifier is wrong
Work you might do
- Defining fill logic that respects depth, queue position, and capacity
- Reviewing agent runs where the analysis was sound and the execution was not
- Designing strategies that pass our grader without doing the job — and telling us
- Formalising what a realistic fill looks like under stress
03Clearing & Settlement LeadClearing, settlement, or middle office
You will encode the constraints that separate a position that looks closed from one that actually settled. Conservation of funds, no overdraft, suspense flat at close — and what it looks like when a system satisfies the report while violating the substance.
What we are looking for
- Operations depth — Current or former clearing, settlement, or middle-office lead at a bank, broker, exchange, or trading firm
- Failure literacy — You know the ways a book balances while the work is wrong: unallocated differences parked in suspense, unsettled cash treated as available, netting that hides a break
- Precision — Comfort specifying a rule tightly enough that code can check it
- Integrity — Willingness to tell us when our verifier is wrong
Work you might do
- Defining the invariants a settlement run must satisfy at every step
- Reviewing agent runs and identifying where the ledger balanced but the work did not
- Designing degenerate strategies that pass our grader without doing the job
- Formalising what settled correctly means
04Quantitative ResearcherHedge fund or bank quant, PhD preferred
You will design the tasks that separate a factor with predictive power from a factor that only looked good on the sample it was fitted to.
What we are looking for
- Research track record — Current or former quant researcher at a fund, bank, or market maker
- Overfitting instinct — You can tell a discovery from a fit in someone else’s backtest
- Statistical rigour — Comfort with out-of-sample design, decay, and multiple-testing correction
- Clarity — Ability to state why a result should not be trusted
Work you might do
- Designing factor-mining tasks with honest holdout structure
- Reviewing model-generated research for silent lookahead and survivorship
- Defining what a real out-of-sample decay curve should look like
- Building degenerate strategies that beat a naive baseline without discovering anything
05Derivatives StructurerStructuring or exotics desk
You will define what makes a payoff correctly specified, correctly hedged, and correctly priced — and where a model output can be internally consistent and still unsellable.
What we are looking for
- Structuring experience — Exotics, structured products, or solutions desk
- Pricing judgment — You know where a model breaks before the market tells you
- Documentation discipline — Term sheets that survive legal and risk review
- Directness — Willingness to say a price is wrong and why
Work you might do
- Defining payoff and hedge constraints as checkable rules
- Reviewing model-generated structures for hidden basis and gap risk
- Designing scenarios where a naive price looks reasonable and is not
- Formalising what a complete term sheet requires
06Risk ManagerBuy-side, sell-side, or exchange risk
You will define the limits that must hold at every step, not just at the close — position caps, VaR, drawdown, and the ways a book passes end-of-day while breaching intraday.
What we are looking for
- Risk ownership — Current or former risk manager with sign-off responsibility
- Intraday judgment — You know the difference between a limit that held and a limit that was never tested
- Quantitative comfort — VaR, stress, and exposure decomposition
- Directness — Willingness to flag a control that looks fine and is not
Work you might do
- Specifying step-level limit checks rather than end-of-period ones
- Reviewing agent runs for breaches that were netted away
- Designing stress scenarios that reveal hidden concentration
- Formalising what an acceptable risk trajectory looks like
07ForecasterTop-ranked on Metaculus, Good Judgment or a prediction market
You will set the bar for forecasts that are scored by the world. Base rates, update discipline and calibration on live events that resolve after the question is asked.
What we are looking for
- Calibration — A public forecasting record you can link to
- Base rates — You start from reference classes, not stories
- Update discipline — You move on evidence, and you can say how much
- Clarity — You write questions that resolve without argument
Work you might do
- Writing forecasting questions with unambiguous resolution
- Setting human baselines on live event feeds
- Reviewing agent forecasts for calibration, not just accuracy
- Designing questions a model cannot answer from memory
T2Licensed & advisory professionsTop rates $120–200 / hr
01Investment BankerBulge bracket or equivalent
You will translate real dealmaking rigour into tasks a model can be graded against — multi-constraint scenarios, client-ready deliverables, and the reasoning behind a defensible number.
What we are looking for
- Banking background — Current or former banker at a top-tier firm with M&A or capital markets exposure
- Analytical communication — Tight comps, defensible assumptions, a clear recommendation path
- Process discipline — Methodical QA and source hygiene
- Judgment — You know which analysis is technically right and commercially useless
Work you might do
- Designing valuation and financing scenarios with transparent assumptions
- Reviewing model output for mis-specified drivers and inconsistent cash-flow logic
- Writing exemplar memos that reflect how bankers actually reason
- Formalising what a client-ready deliverable requires
02Management ConsultantMcKinsey, Bain, BCG, or equivalent
You will encode the thinking systems behind elite consulting — framing ambiguous problems, developing hypotheses, pressure-testing data, and distinguishing insight from restatement.
What we are looking for
- Consulting pedigree — Partner, Principal, or Senior Engagement Manager at a top strategy firm
- Structured problem-solving — Hypothesis-driven decomposition of cross-functional problems
- Executive communication — Board-level synthesis and storyline architecture
- Quality bar — A clear sense of what great looks like in client-ready output
Work you might do
- Rewriting model analyses into consulting-grade frameworks
- Designing evaluation tasks that capture partner-level reasoning under constraint
- Reviewing outputs for logic gaps and synthesis quality
- Formalising what a defensible recommendation requires
03LawyerLitigation, regulatory, or transactional
You will bring interpretive judgment to work where the answer depends on precedent, drafting, and what a counterparty could argue. Where a document can be technically compliant and commercially dangerous.
What we are looking for
- Legal credentials — Practising or former counsel at a firm, in-house team, government office, or court
- Interpretive precision — Parsing argument structure and identifying hidden assumptions
- Drafting mastery — Documents that survive adversarial reading
- Integrity — Commitment to principled reasoning over convenient reasoning
Work you might do
- Reviewing model reasoning for logical and interpretive accuracy
- Designing tasks where the compliant answer and the correct answer differ
- Building structured argumentation datasets
- Formalising what a defensible legal analysis requires
04Tax AttorneyJD with an LL.M. in tax, or a CPA in tax practice
You will define when a tax position is defensible and when it only looks tidy. Filings, elections and the documentation that has to exist before the number does.
What we are looking for
- Licence — Admitted attorney or CPA in active tax practice
- Depth — Corporate, cross-border or indirect tax experience
- Evidence — You know which document proves a position
- Precision — You can state a rule a verifier can apply
Work you might do
- Writing tax scenarios with hidden traps and reference answers
- Reviewing agent filings for positions that would not survive audit
- Encoding documentation requirements as checkable rules
- Flagging answers that read correctly and would cost the client
05Controller / CPAHas closed real books
You will define the ways a set of books can balance and still be wrong. Differences parked in suspense, adjustments without evidence, reversals that net to nothing.
What we are looking for
- Close experience — You have owned a period-end close, not just reviewed one
- Failure literacy — You know which shortcuts produce a clean trial balance and a dirty ledger
- Evidence discipline — Every adjustment ties to a statement line or a voucher
- Directness — Willingness to say a book is wrong when it balances
Work you might do
- Defining the invariants a close must satisfy at every step
- Reviewing agent runs for adjustments without supporting evidence
- Designing reconciliation tasks with planted, realistic discrepancies
- Formalising what closed correctly means
06Internal AuditorBig Four or in-house audit
You will encode the red flags: split invoices under an approval threshold, self-approval, approval timestamps after execution, three-way mismatches that still paid.
What we are looking for
- Audit experience — Big Four, internal audit, or regulatory examination background
- Control literacy — You know which controls pass while the risk remains
- Pattern recognition — Splitting, timing, and routing schemes
- Precision — Comfort turning a red flag into a machine-checkable rule
Work you might do
- Defining detection rules for threshold evasion and retroactive approval
- Reviewing agent workflow runs for control bypasses
- Designing scenarios where the audit trail looks complete and is not
- Formalising segregation-of-duties constraints
07ActuaryQualified, insurance or pensions
You will define where a reserve, a pricing assumption, or a projection is defensible — and where a number can be arithmetically correct and professionally indefensible.
What we are looking for
- Qualification — Fellow or Associate of a recognised actuarial body
- Assumption discipline — You know which assumptions drive the answer and which are decoration
- Regulatory awareness — Reserving and solvency standards
- Clarity — Ability to state why a projection should not be relied on
Work you might do
- Defining assumption-validity checks for model-generated projections
- Reviewing reserving work for unsupported judgment
- Designing scenarios with realistic tail behaviour
- Formalising what a defensible actuarial opinion requires
08PhysicianLicensed, any specialty
You will define where a clinically plausible answer is unsafe — the distinction between an output that reads correctly and one a clinician would act on.
What we are looking for
- Licensure — Currently or formerly licensed to practise
- Clinical judgment — Comfort with differential reasoning and uncertainty
- Safety literacy — You know which errors are recoverable and which are not
- Precision — Willingness to specify exactly why an output is unsafe
Work you might do
- Reviewing model reasoning for clinically dangerous confidence
- Designing cases where the plausible answer is the wrong one
- Defining what a safe recommendation requires
- Flagging outputs that satisfy a rubric and fail a patient
09Nurse PractitionerBoard-certified NP, two or more years in practice
You will judge clinical plans the way a patient experiences them. Triage, medication, follow-up and the plans that read well but would cause harm.
What we are looking for
- Licence — Active NP licence with national board certification
- Practice — Two or more years of clinical work after certification
- Clinical reasoning — You can write down why, not only what
- Safety — You spot the plausible plan that would hurt the patient
Work you might do
- Writing realistic patient scenarios and decision pathways
- Creating reference answers with the reasoning behind them
- Reviewing agent plans for diagnostic accuracy and safety
- Designing cases where the obvious answer is dangerous
10PharmacistLicensed PharmD, community or hospital
You will define what a safe dispensing decision looks like. Interactions, dosing, substitutions and the checks that stop an error before it reaches a patient.
What we are looking for
- Licence — Active pharmacist licence
- Practice — Community, hospital or clinical pharmacy experience
- Interaction knowledge — You catch the combination others miss
- Precision — You can express a check as a rule
Work you might do
- Writing medication scenarios with interaction and dosing traps
- Reviewing agent recommendations for safety and completeness
- Encoding dispensing checks a verifier can apply
- Flagging answers that are correct in general and wrong for this patient
T3Advanced STEM & reasoningTop rates $130–200 / hr
01MathematicianPhD or equivalent
You will design problems where the final answer is verifiable and the path to it is not — and define what separates a correct derivation from a correct guess.
What we are looking for
- Academic or research depth — PhD in mathematics, or equivalent research output
- Proof discipline — You can tell a proof from a plausible argument
- Problem design — Ability to construct problems with unambiguous ground truth
- Clarity — Comfort writing for an intelligent non-specialist
Work you might do
- Designing problems where the obvious method fails
- Reviewing model derivations for gaps that do not change the answer
- Defining step-level correctness criteria
- Building problem sets with controlled difficulty
02Competition MathematicianIMO, Putnam or national olympiad medallist
You will write and judge problems that separate reasoning from pattern matching. Hard puzzles with one checkable answer and no shortcut.
What we are looking for
- Record — Olympiad or Putnam results, or published problem sets
- Taste — You know when a problem is hard for the right reason
- Rigour — Every answer you write can be checked exactly
- Adversarial instinct — You look for the shortcut before a model does
Work you might do
- Writing original puzzles with verified unique answers
- Reviewing solutions for reasoning, not just the final number
- Attacking the answer checker with degenerate submissions
- Retiring problems once they leak or saturate
03Operations Research ScientistPhD in OR or optimisation, or production solver experience
You will define what a proven solution means for matching, scheduling, packing and allocation. A feasible answer, or a certificate that none exists.
What we are looking for
- Training — PhD in operations research, optimisation or equivalent practice
- Solvers — Experience with MIP, CP or custom heuristics in production
- Proof — You can tell a certificate from a claim
- Precision — You write constraints a checker can verify in milliseconds
Work you might do
- Designing optimisation families with generated instances
- Writing feasibility and infeasibility certificates
- Reviewing agent solutions for hidden constraint violations
- Finding instances where a greedy answer looks optimal
04PhysicistPhD or equivalent
You will define where a model’s physical reasoning is dimensionally correct and physically impossible, and design tasks that separate the two.
What we are looking for
- Research depth — PhD in physics, or equivalent research output
- Physical intuition — You notice when a result violates something no equation stated
- Modelling judgment — Comfort with approximation and its limits
- Precision — Ability to specify what makes a result valid
Work you might do
- Designing problems with realistic constraints and unambiguous answers
- Reviewing model reasoning for unphysical intermediate steps
- Defining validity checks beyond the final number
- Building graded problem sets across subfields
05ChemistPhD or equivalent
You will define where a proposed route, mechanism, or condition is theoretically sound and practically impossible.
What we are looking for
- Research depth — PhD in chemistry, or equivalent industrial experience
- Bench judgment — You know which reactions work on paper and not in a flask
- Safety awareness — Comfort flagging hazardous or infeasible proposals
- Precision — Ability to specify exactly why a route fails
Work you might do
- Reviewing model-proposed syntheses for practical feasibility
- Designing tasks where the textbook answer is wrong
- Defining validity criteria for mechanisms and conditions
- Building problem sets with verifiable outcomes
06Life ScientistPhD or equivalent
You will define where a biological claim is supported, where it is extrapolated, and where a model has produced a plausible sentence with no evidential basis.
What we are looking for
- Research depth — PhD in a life science, or equivalent research output
- Evidence discipline — You distinguish a finding from a hypothesis
- Method literacy — Comfort assessing experimental design and statistics
- Clarity — Ability to state what a study does not show
Work you might do
- Reviewing model reasoning for unsupported biological claims
- Designing analysis tasks with verifiable conclusions
- Defining criteria for adequate evidence
- Flagging conclusions that outrun the data
07University ProfessorTop-ranked institution
You will help transform raw model behaviour into structured, teachable insight — writing and refining reasoning examples that reflect how experts actually argue.
What we are looking for
- Academic standing — Current or former professor at a top-ranked institution, with published research
- Scholarly depth — Advanced domain knowledge and the ability to explain it plainly
- Analytical writing — Tightly reasoned, well-structured prose
- Precision — A methodical approach to critique and review
Work you might do
- Reviewing model responses for sound argumentation and conceptual clarity
- Drafting exemplar answers that illustrate expert reasoning
- Conducting comparative evaluations to surface subtle reasoning flaws
- Designing datasets that capture authentic expert cognition
T4Engineering & machine learningTop rates $120–220 / hr
01Post-Training ResearcherFrontier lab or equivalent
You will design reward functions and then try to break them. This is the work the labs say they are short of: writing a grader, attacking it with a model that is actively trying to cheat, and confirming afterwards that it measured what you intended.
What we are looking for
- Post-training experience — RLHF, RLVR, or reward-model work at a lab or serious research group
- Adversarial mindset — You reach for the shortcut before you reach for the solution
- Systems judgment — Understanding of how a weak reward propagates through a training run
- Rigour — Comfort with variance, pass^k, and contamination checks
Work you might do
- Designing invariant-based reward functions for multi-step tasks
- Running degenerate policies against graders and documenting the bypasses
- Reviewing reward specifications for exploitable slack
- Formalising what a robust verifier looks like in a new domain
02ML Research ExpertPublished author with benchmark or competition work
You will judge whether an agent did machine-learning research or only produced a score. Data handling, experiment design and results that hold on held-out answers.
What we are looking for
- Publications — First-author papers or top competition finishes
- Experiment design — You know what a fair baseline is
- Leakage instinct — You find the split that leaked
- Clarity — You can write a task an agent cannot misread
Work you might do
- Authoring research tasks graded by the platform that holds the answers
- Reviewing agent experiments for leakage and cherry-picking
- Writing reference solutions and baselines
- Designing tasks where overfitting the public split fails
03Machine Learning EngineerKaggle Grandmaster or Master preferred
You will identify the gap between a validation score and a real one. Leakage, split design, and the distance between what a model reports about itself and what the holdout says.
What we are looking for
- Competition record — Kaggle Grandmaster or Master, or equivalent applied track record
- Leakage instinct — You find the target buried in a feature before you trust the score
- Split discipline — Time-aware, group-aware, and distribution-shift-aware validation
- Honesty — Comfort reporting that your own result did not hold
Work you might do
- Designing tasks where the obvious validation strategy is the wrong one
- Reviewing agent runs for leakage, tampered folds, and unfixed seeds
- Measuring the gap between self-reported CV and official holdout
- Building submissions that score well for the wrong reasons
04Open-Source MaintainerMaintainer of a widely used open-source project
You will decide which fixes survive the next release. Code review with the codebase’s future in mind: API drift, deprecations and patches that only pass today’s tests.
What we are looking for
- Maintainership — Merge rights on a project people depend on
- Review judgment — You reject patches that pass and still rot
- History — You know how codebases change release to release
- Precision — You can write a test that encodes the concern
Work you might do
- Reviewing agent patches against simulated future versions
- Writing tests that separate lasting fixes from short-sighted ones
- Curating tasks verified to distinguish good fixes from bad
- Flagging patches a maintainer would never merge
05Software EngineerProduction experience, any stack
You will review code an agent wrote and decide whether it works, whether it would survive review, and whether the tests that passed actually tested anything.
What we are looking for
- Shipping experience — You have maintained code other people depended on
- Review judgment — You can tell a passing test from a meaningful one
- Debugging skill — Finding the bug rather than the symptom
- Clarity — Writing a review comment that explains the failure
Work you might do
- Reviewing and ranking model-generated solutions with technical rationale
- Writing reference implementations a model should have produced
- Designing tests that catch the edge cases a model missed
- Flagging patches that pass CI and break production
06Data ScientistApplied, production experience
You will build the tasks where feature engineering quietly decides the outcome, and define what separates a legitimate feature from one that will not exist at inference time.
What we are looking for
- Production experience — You have shipped a model other people depended on
- Feature discipline — Clear judgment on what is available at prediction time
- Diagnostic skill — Distribution shift, drift, and silent degradation
- Clarity — Ability to explain why a metric improved for the wrong reason
Work you might do
- Designing datasets with deliberately planted leakage
- Reviewing agent feature engineering for future information
- Measuring distribution-shift sensitivity across tasks
- Writing the rationale behind each planted trap
T5OperationsTop rates $80–140 / hr
01Treasury OperationsCorporate or bank treasury
You will define what makes cash actually available — settlement timing, nostro balances, and the difference between a balance on a screen and money you can move.
What we are looking for
- Treasury experience — Cash management, liquidity, or funding operations
- Settlement literacy — T+N mechanics and the cost of assuming otherwise
- Operational judgment — You have handled a funding gap in real time
- Precision — Comfort specifying availability rules exactly
Work you might do
- Defining availability rules that respect settlement cycles
- Reviewing agent runs where unsettled cash was treated as available
- Designing liquidity-constrained scenarios with real funding pressure
- Formalising nostro and suspense reconciliation
02Supply Chain PlannerManufacturing, distribution, or 3PL
You will define the constraints that make a plan executable rather than merely feasible on paper — lead times, capacity, and the ways a schedule satisfies the model and fails in the warehouse.
What we are looking for
- Planning experience — S&OP, production planning, or distribution planning ownership
- Constraint realism — You know which constraints are soft in a spreadsheet and hard in a building
- Systems familiarity — ERP or planning system experience
- Clarity — Ability to state why a plan will not survive contact
Work you might do
- Defining executable-plan constraints for agent scheduling tasks
- Reviewing agent plans for violations a naive checker would pass
- Designing disruption scenarios with realistic recovery options
- Formalising what a feasible plan means
03Customs & Trade ComplianceLicensed broker or in-house
You will encode classification and duty logic — where the rules are objective, where they are contested, and where a defensible answer differs from a correct one.
What we are looking for
- Classification experience — Licensed customs broker or in-house trade compliance
- Regulatory currency — You track how the rules changed this year
- Judgment under ambiguity — Comfort with contested classifications
- Documentation discipline — Every determination has a cited basis
Work you might do
- Defining classification tasks with defensible ground truth
- Reviewing agent determinations for unsupported reasoning
- Designing scenarios where the obvious classification is wrong
- Formalising duty and origin calculation checks
04Cross-border E-commerce OperationsShopee, Lazada, or equivalent
You will define the workflows on platforms few AI environments cover, the ones many of Asia’s sellers open every morning.
What we are looking for
- Platform experience — Hands-on seller or ops experience on non-US marketplaces
- Process knowledge — Listing, pricing, fulfilment, dispute, and returns flows
- Local fluency — You know the rules that are never written in English
- Precision — Comfort turning a habit into a specification
Work you might do
- Documenting real platform workflows step by step
- Reviewing agent runs for actions that are technically valid and operationally wrong
- Designing tasks with realistic platform constraints
- Formalising what a completed listing or dispute looks like
05Travel & Itinerary OperationsCorporate travel, DMC, or tour ops
You will define the scheduling constraints that hold in practice and not on paper — last admission times, weekly closures, queue times, and the buffers a plan needs to survive one delay.
What we are looking for
- Operations experience — Corporate travel, destination management, or tour operations
- Local knowledge — Asian markets preferred: closing days, reservation windows, transit realities
- Constraint realism — You know which itineraries look fine and fall apart on the day
- Precision — Comfort turning local practice into an explicit rule
Work you might do
- Defining schedule constraints that chain: last admission, queue, and transit together
- Reviewing agent itineraries for plans feasible only on paper
- Designing tasks with realistic closure and reservation rules
- Formalising what a workable itinerary means
T6General & writingTop rates $40–70 / hr
01Technical WriterPublished documentation or API reference work
You will make tasks and rubrics impossible to misread. Clear instructions are the difference between a model failing the task and failing to understand it.
What we are looking for
- Portfolio — Documentation, API references or procedures you wrote
- Clarity — One reading, one interpretation
- Structure — You organise information for someone in a hurry
- Care — You notice the ambiguous word
Work you might do
- Editing task instructions and rubrics for ambiguity
- Writing reference answers in plain language
- Reviewing agent explanations for accuracy and completeness
- Flagging instructions that two experts would read differently
02Generalist ReviewerDegree-level, fluent English
You will review agent runs and flag where the output looks right and the work was not done. No domain licence required — careful reading and clear writing are the qualification.
What we are looking for
- Careful reading — You notice when a conclusion does not follow from what came before
- Clear writing — Complete sentences, specific objections
- Consistency — Comfort applying the same standard across many runs
- Honesty — Willingness to say you are unsure
Work you might do
- Reviewing agent trajectories against a written standard
- Flagging outputs that satisfy the format and miss the task
- Writing short, specific notes on why a run failed
- Helping calibrate difficulty across a task set
03Multilingual SpecialistJapanese, Korean, Thai, Indonesian
You will work on regional software that millions of people use every day and few AI environments cover, such as Shopee and Lazada.
What we are looking for
- Native fluency — In at least one of the listed languages, plus working English
- Platform familiarity — Real use of the local enterprise or commerce tools
- Cultural precision — You catch what a translation loses
- Care — Comfort with detailed, repetitive review
Work you might do
- Documenting workflows on non-US enterprise software
- Reviewing agent runs for locale-specific failures
- Translating task specifications without losing constraints
- Flagging conventions a non-local reviewer would miss
Questions
What experts ask before joining
How much will I get paid?
Each role shows its top rate, from up to $40 an hour for generalist review to up to $300 an hour for the most specialised markets work. Your starting rate depends on the credentials you bring; sustained quality moves you towards the top rate.
Do I need a specific credential?
It depends on the tier. Licensed professions and advanced STEM roles require the credential named in the listing. Operations and generalist roles weigh demonstrated experience as heavily as formal qualification.
How flexible is the work?
Fully. No minimum hours, no fixed schedule, no standups. You choose which tasks to take. Everything is remote and contract.
What does the work actually involve?
Defining what correct means in your field, reviewing agent runs against that definition, and designing cases where the obvious answer is wrong. You are writing the rules a model gets graded against — not labelling data.
How long does the application take?
About a minute. One role, one line of background, one link, and one question. We read the question.
When will I hear back?
We work down the list by role. Tiers with the smallest pools move first. If your background matches an active environment, you will hear from us directly.
Who is the work for?
Frontier AI labs and vertical AI companies building agents for work where being approximately right is the same as being wrong. We have no model-company shareholders.
Help us build worlds that answer back
Apply for a specific role. We read every answer, and reply to the ones that fit an open world.
