Remote AI training and data labeling jobs

Write realistic feature and bug-fix tasks for AI coding models, plus deterministic verifiers that accept valid solutions and reject the rest. About 15 flexible hours a week, paid per accepted task on a $30–100/hour band, 100 openings; a strong testing background is the stated plus.

Write and review Lean 4 proofs for a leading AI lab, formalize informal mathematics and judge whether a model's proof actually proves the right statement. A W-2 part-time employment position through Cincinnatus LLC, at least 20 hours a week and up to 40, paying $90–110/hour.

Embedded firmware, FPGA/RTL, flight software and hardware test automation engineers review and write hard problems about software that runs on real hardware (timing, race conditions, interfaces, bring-up) for an AI research initiative. 5+ years, 3 recent hands-on. Remote contract, $100–120/hour.

Maintainers of real open source repositories review LLM-generated patches against their own codebase, justify each verdict in writing, and peer-review other experts. Open globally, about 15 hours a week, paid per accepted task at a stated $150–300/hour equivalent. 10 openings.

Build benchmark tasks for Terminus Bench 4 from real terminal and infrastructure problems: a problem setup, a working environment, a reference solution and an automated verifier for each. Remote independent contractor work through Meridial at $35/hour.

Experienced IT and systems managers: critique IT strategy, architecture, vendor and security-compliance scenarios and write the expert input that trains AI tools for IT professionals. 5+ years leading technical teams or enterprise infrastructure. 5 openings, contractor, $60–75/hour.

Work through real Stripe workflows (customers, payments, invoices, subscriptions, refunds, disputes) while recording your screen, then write prompts and rubrics from what you did. Needs current Stripe access; prior human data work preferred. Forty openings, $90–170/hour.

Remote hourly contract for Android developers who already run Android Studio on their own Mac. $55–65/hour, paid weekly via Stripe or Wise. The software is free; you need a Mac with a display above 2.5 megapixels (MacBook Pro 14" and 16" screens qualify). Tasks are not described.

Remote hourly contract for Mac power users who run macOS Sonoma 14.5 as their everyday OS. $60–70/hour, paid weekly via Stripe or Wise. The version is specific, so anyone who has upgraded past Sonoma may not qualify. MacBook Pro 14" and 16" screens meet the display requirement.

Despite the title, a CAD and computational-geometry role: turn dimensioned engineering drawings into validated 3D solids built in Python (build123d, CadQuery and similar) and exported as STEP, for an AI training project. Open globally, about 15 hours a week, per-task pay. 10 openings, $50–100/hour.

Competitive programmers write original problems absent from existing online judges, with a reference solution, a robust test suite and C++ checkers for multi-answer outputs, as RL environments for AI. 15+ hours a week, paid per accepted task against a $45–65/hour band. 300 openings.

Fix bugs, build features and refactor real codebases in Python, Java, Rust, C++, Go or TypeScript, then write coding tasks and feedback that test AI models. Competitive programming and open-source experience wanted. Paid per accepted task, $50–100/hour band, 100 openings, remote contractor.

A short, well-paid sprint for very senior software engineers: help a leading foundation-model lab improve its models on hard SWE tasks. 10+ years at top US tech firms, about 20 hours a week for 2–3 weeks. $150–210/hour by geography and level; the 2-hour vetting exercise is paid $100.

Work through real Google Analytics (ideally GA4) tasks on screen recording, then write prompts and rubrics from them so AI agents learn the workflows. Needs your own current GA access and prior human-data work. 40 openings, contractor, $30–100/hour.

Design GPU programming tasks in CUDA, WebGPU or GLSL for training LLMs on performance and architecture, including kernel profiling and C++ host code. Remote contractor, $60–95/hour paid per accepted task, 25 openings. Graphics, HPC and ML acceleration backgrounds all qualify.

Remote hourly contract for people who run Ubuntu 24.04 as their everyday desktop OS: Linux users, sysadmins, DevOps and open-source developers. $60–70/hour, paid weekly via Stripe or Wise. No licence cost, but you need a display above 2.5 megapixels. Tasks are not described.

Create, solve and review hard full-stack engineering tasks, debug across browser, API and database layers, judge AI-generated solutions, and port whole builds from one language to another. Global and fully remote, about 15 flexible hours a week, paid per task on a $50–100/hour band, 300 openings.

A paid expert conversation for engineers who build and run LLM agents in production: a short AI screening interview (no coding), then, if selected, a 30-minute live call on agent reliability, evaluation and internal adoption, paid $100–500 depending on depth of experience.

Sales ops and CRM people: carry out real HubSpot workflows (contacts, deals, pipelines, automation) on screen recording, then write the prompts and rubrics AI agents are trained and judged on. Needs your own HubSpot access. 40 openings, contractor, $28–92/hour.

Security-minded engineers debug vulnerabilities, audit codebases, run penetration tests and refactor code in Python, Java, Rust, Go, C++ or TypeScript to produce AI training data. About 15 hours a week, flexible, paid per accepted task against a $30–100/hour band. 300 openings.

Expert C# and .NET developers with strong MySQL skills build and maintain backend services and APIs on a micro1 project framed as AI training. Remote contract, only 2 openings, $60–110/hour.

Backend engineers, ideally with an AppSec background, write realistic coding tasks for AI (features and bug fixes) plus deterministic verifiers that accept valid solutions and reject the rest. About 15 hours a week, flexible schedule, paid per accepted task against a $30–100/hour band. 300 openings.

Build reinforcement learning environments where an AI agent must fix bugs, add features or optimise code using real MCP servers, with deterministic verifiers and golden solutions. About 15 hours a week, your own schedule, paid per accepted task against a $60–120/hour band. 100 openings.

Contribute real-world code, explanations and code reviews as AI training data, drawing on five-plus years in Python, Java, JavaScript/Node.js or full-stack work. A narrow $60–75/hour band, 15 openings, remote contract with documentation and architecture writing at its core.