STARTINGIN AIWORK✳
For people new to data annotation and AI evaluation. Read what the work involves, learn the basics and get advice from someone who does this work.
Where to begin.
Most beginners piece this field together from scattered forum posts. This route keeps it to four steps. The guides are free and optional, they have no tests, and you can message me for a person's view. Not sure where you fit? Answer five short questions ↗
- First
Understand the work
What AI workers do, how contract work differs from a regular job and what the screening looks like.
Read the beginner's guide ↗ - Then
Learn one skill
Start with labeling data or with judging model answers. Both guides include practice examples.
Data annotation ↗Evaluating responses ↗ - Next
Prepare for screening
How platforms test new people, how to explain your reasoning and how to present your skills honestly.
Your first AI interview ↗Your first CV ↗ - Finally
Apply on the platform
Look at the types of projects, check that the platform accepts people from your country and apply on its own site.
Types of AI projects ↗
I built this site to share the basics I wish I had known before my first application.
Denys HaidamakaBeginner
guides.
Short guides for people who have never done an AI task. You don't need an account, and there are no quizzes.
Data annotation
Labeling images, classifying text, reviewing audio and why accuracy matters.
Read guide ↗Evaluating model responses
Comparing answers, checking instructions and giving feedback based on evidence.
Read guide ↗Your first CV
How to present transferable skills honestly.
Resume guide ↗Another guide for reviewers, Spotting AI-isms in writing ↗
How a language model
writes an answer.
This lesson follows one request from prompt to answer, one token at a time.
A simplified teaching model of a large language model.
The 3D model loads when this section comes into view. The explanations work without it.From prompt to answer.
You write a prompt. The model turns it into tokens, works through them with its context, predicts the next token, adds it, and repeats until the answer is complete.
“Explain data annotation in one short sentence for someone new to AI work.”
The lesson follows this request through every step.Knowing the steps helps you judge the answer itself. Does it meet the request with the information the model had?
The model adds each chosen token to the text before it predicts the next one. A stop token ends the answer. The 3D model runs the same loop.
After “Data annotation means …”, the model gives every possible next token a probability. Temperature is a setting of the product that runs the model. It controls how evenly the choice spreads across the candidates.
Inside every block, each token’s representation draws on earlier tokens. Select a word to see which earlier words it uses most.
The model can only use text that fits in its context window. This one holds 16 tokens. Real windows hold thousands to over a million.
The model writes its answer with fixed weights, the adjustable numbers inside the blocks. Training set them earlier. In a chat they stay the same, and the model only uses its context.
- Pre-training
Learning to predict text
Training on a very large collection of text adjusts the weights step by step until next-token predictions improve. The model gets its general language patterns and knowledge here, up to a cut-off date.
- Fine-tuning
Learning to follow instructions
Smaller sets of example tasks with good answers teach the model to follow requests. People often write or check these answers.
- Human feedback
Learning from people’s judgements
People compare and rate answers, and good ratings matter as much as bad ones. Developers use these ratings to adjust the model further, for example with reinforcement learning from human feedback (RLHF), and to evaluate it.
Where AI work fits. Data annotation and evaluation tasks can feed fine-tuning and human feedback, or test the finished model. Some tasks never reach training. Your project instructions explain how the team uses your work.
Some products offer “memory”. It usually stores text and adds it to the context later. It does not retrain the weights.
These checks apply to a finished response. You can't trace them to one part of the model. Your project's own definitions and instructions come first.
Does it do what the user asked?
A polished response can still miss, partly meet or contradict a requirement such as length, format, audience or scope.
Example. The request asks for one short sentence, and the answer gives five paragraphs.
How to check. Compare the answer with each requirement in turn, and follow your project's rules on conflicts and exceptions.
Read the AI evaluation guide ↗A simplified teaching model. The board, plates and blocks stand for computations, and real models have many more than three layers. A language model writes text one token at a time from its prompt and context. Its context has a limit, and fluent answers still need evaluation. The example on this page follows a script, and I made up the numbers and token splits in the panels for teaching.
Some projects require specialized knowledge or advanced English. Not every task is entry-level, and available work changes. The guides explain the differences.
FACT
CHECKING
BASICS✳
How to trace claims to their original sources, check dates and evidence, and search with Google operators. The guide has no course or exam.
Open fact checking labsite:who.int "report" filetype:pdfScreenings and
interviews.
Most platforms run their own screening. These guides help you prepare without another formal test.
Your first AI
interview & assessment
What the screening may involve, how to think aloud and how to avoid rushing.
Open guide ↗Expert assessment
practice
How to turn professional knowledge into clear answers you can defend.
Open guide ↗Quality review
scenarios
Applying a rubric consistently, judging severity and writing clear feedback.
Open guide ↗Types of
AI projects.
These are the main kinds of projects beginners take on. I will add real platform details and referral links here later. A listed platform is no promise of an opening.
These categories are previews. They are neither live vacancies nor partnerships. Later, each verified link will open the platform's official application page.
Want to ask a real person? ↗Questions and
personal advice.
If you're new to this field and don't know where to begin, tell me about your experience. When I have time, I'll suggest where to start.
You share what you already know and want to do
I review your background when time allows
You get practical suggestions with no promise of a job
Your message
A short message is enough.
I'll open the form here once I set up secure sending. Until then, LinkedIn is the quickest way to reach me.
- Say what you have worked on and which languages you use.
- Tell me which kind of AI work you are curious about.
- Ask one concrete question.
You don't need a CV for a first message. Please don't send ID documents or confidential project material. LinkedIn's own terms apply there.
AI for your business.
I also help businesses find where AI and automation can simplify daily work, from repetitive admin tasks to agent-assisted workflows with clear human oversight.
Explore AI for BusinessAbout
me.
I'm Denys Haidamaka. I've worked on AI evaluation, multimodal review, quality checks and evaluation frameworks. I know the questions people have when they first enter this field.
FacinateAI Hub is my independent project to share practical, beginner-friendly advice and point people toward opportunities. I don't represent, hire for or speak on behalf of the platforms featured here.
Start a conversation ↗Frequently asked
questions.
Do I need programming experience to start? +
Not always. Some annotation, data labeling and evaluation tasks focus on language, attention to detail and following instructions. Specialized projects may require professional knowledge, coding or other qualifications.
Is AI work a guaranteed remote salary? +
No. Many opportunities are contract-based and depend on country, eligibility, screening results and project demand. Available work and pay can change.
Will you help me pass the platform's assessment? +
I share general preparation tips and original practice examples. I don't provide confidential test answers, guarantees or shortcuts around platform rules.
Do I have to send you my CV? +
No. You can read every guide and apply using a referral link without sending me any personal details. Asking for advice and sharing a resume are optional.
Do you receive referral compensation? +
Some links may earn me a reward if you meet the platform's rules. That doesn't mean the platform will accept or pay you. I recommend platforms by fit, and each eligible link shows the details.
Can I contact you through this site? +
Not yet. The site needs secure handling for messages and CVs first, and it never sends or stores your data in the background. Until then, message me on LinkedIn.