현재 Alignerr에서 진행 중인 6가지 AI 교육 프로젝트 [2026년 3월]

Most people apply to Alignerr for a generic title like "AI Training Specialist" or "Data Labeling Specialist." What they don't realize is that those are just the front door. Once you are inside, the real work happens on named projects — each with a specific discipline, skill requirement, and pay structure. Right now there are six active projects running on Alignerr. Some are open to anyone. Some require niche credentials. And a few — like Code Human and Project Prism — are only offered to the platform's most qualified contributors. This guide breaks down each project: what the work actually involves, who it is designed for, and how to apply directly. How Alignerr Projects Work Alignerr is the talent network built on top of Labelbox, the data infrastructure platform. When you apply, you go through a screening and a Zara video interview. After that, you are not automatically given work — you are placed in a pool and matched to active projects based on your skills and assessment scores. Important: The Assessment Comes First You cannot jump straight into a named project. All contributors must first pass the Alignerr assessment and Zara video interview before being eligible for any project. If you have not started yet, use the referral links below — they are specific to each project. | Project | Domain | Pay | Difficulty | |---|---|---|---| | Gamechanger — Lua Project | Game Dev / Coding | $40–$120/hr | Intermediate | | ATC Transcription Project | Aviation / Audio | $20–$30/hr | Specialist | | Project Prism | AI Evaluation | $40–$120/hr | Advanced | | Code Human Project | Software Engineering | $40–$120/hr | Advanced | | Project Rainforest | Voice Acting | Per approved task | Beginner | | PR Writer w/ Feedback Project | Code Review | $40–$120/hr | Advanced | Gamechanger — Lua Project This project targets Lua developers and game programmers. Your job is to build real gameplay systems — not toy examples — that help AI learn how experienced game developers write production-quality code. What the Work Involves - → Design and implement gameplay systems in Lua - → Write clean, idiomatic game logic and design systems - → Create unit tests that match how senior devs work - → Demonstrate how interactive, complex elements are built Who Should Apply - ✓ Lua developers (professional or hobbyist) - ✓ Game developers familiar with gameplay systems - ✓ Anyone who can write and test game logic from scratch Note: You don't need a shipped title. You need to be able to write the code. ATC Transcription Project This project is building AI for the aviation industry — specifically, systems that can understand and process real air traffic control (ATC) audio. The work involves transcribing genuine tower-to-pilot recordings with professional-grade accuracy. This one is niche — but that is the point. Alignerr isn't looking for fast typists here. The audio contains heavy accents, background noise, fast-paced phraseology, and aviation-specific callsigns. Only someone with real ATC or aviation experience can do this accurately. What the Work Involves - → Transcribe real ATC-to-pilot audio recordings - → Label speakers and accurately capture callsigns - → Handle aviation-specific abbreviations and phraseology - → Deliver high-accuracy transcripts per formatting guidelines Who Should Apply - ✓ Air traffic controllers (active or retired) - ✓ Pilots, flight instructors, or dispatchers - ✓ Anyone with strong ATC communication experience - ✓ Bonus: prior transcription or annotation experience Project Prism Project Prism is not data annotation. It is adversarial AI evaluation — the job of designing prompts so challenging that state-of-the-art models fail them. If you like poking holes in AI reasoning, this is the project for you. The goal is simple to describe and hard to execute: craft a prompt with a structured rubric where at least two out of three leading AI models cannot meet the standard. Then write a clear, evidence-based justification for why they failed. What a Prism Task Looks Like - You write the prompt. It needs to require specialist knowledge, multi-step reasoning, or both. The goal is to expose genuine limitations — not trick questions. - You write the rubric. Define exactly what a correct answer looks like, with clear, measurable criteria. - You test it against top models. Submit it and verify that at least two SOTA models genuinely fail your rubric. - You write the justification. Explain specifically where and why each model fell short. Who This Is For Code Human Project Code Human is Alignerr's flagship coding evaluation project. Contributors act as senior software engineers — prompting AI models to solve real-world coding problems, comparing outputs, and providing the detailed technical feedback that trains the next generation of coding assistants. Step 1 Prompt the Models Give two competing AI models a real coding task — something from your own work experience Step 2 Compare Outputs Evaluate both responses for correctness, effi