POLARISAI ACADEMY

AI jobs

Most paid AI work falls into a few families, from labeling examples to keeping a model running for other people. This page names those families, shows how to read a job ad, and sets out what a portfolio needs at each level.

AI jobs in Sweden

From Platsbanken and employer sites · read 29 Oct 2026

AI Architect · Knowit · StockholmApply by 27 Mar · Platsbanken
AI Engineer · Norrminds · LinköpingApply by 27 Dec · Platsbanken

Beginner

Four families, in plain words

Most paid AI work falls into four families. Match a job ad to a family by the tasks, and treat the title as a label.

An annotator marks examples so a model can learn. A model is software that finds patterns in examples and then guesses on something new. The annotator opens a photo, a sentence, or a short recording and applies a label from a written rule, such as damaged roof, refund request, or invoice total matches the lines. The skill is the same careful decision on every item. Early weeks in this family seldom require code.

A coordinator runs the queue. They assign the day's batch, settle cases where two people labeled one item differently, and record which version of the rule was in force.

A support or automation role sits inside another job. A person in a workshop, a shop, or a finance desk uses a chat tool, a checklist, or a short script so a repeated task takes fewer steps.

A builder writes the software around a model: the code that sends it data, checks its answers, and puts it in front of users. Most job ads titled AI Engineer belong here.

Swedish official codes make the families easier to match. In SSYK 2012, the Swedish Standard Classification of Occupations, code 3512 is IT support technician. Code 2512, software and systems developers, covers people who analyze, design and develop software, applications, systems and databases. A first portfolio can be fifty labeled examples with the rule at the top, or one script that cleans a spreadsheet, plus a note of what you checked by hand. Higher-vocational courses are listed by the Swedish National Agency for Higher Vocational Education. Employers cannot post their own job ads here yet. That opens later.

Turn one job ad into plain language

You are helping an adult who is new to AI jobs. I will paste a job ad below. Do three things in short sentences. 1) Say which family the tasks belong to: annotator (labels examples from a written rule), coordinator (runs the queue and the rule log), support or automation (speeds up an existing job), or builder (writes software). 2) List the duties, and list the tool names separately. 3) Say what a beginner could finish in a weekend as proof for this job ad. If the task list is missing, say what is missing. Do not invent the employer or the pay. Job ad: [paste the job ad here]

Intermediate

Read the duties, then the pay

Read the duties in a job ad before the title. Ask which family the work belongs to, which SSYK or O*NET code fits, and which lines are tools.

Data scientists in the United States are O*NET 15-2051.00. The official summary has them turn raw data into information with programming and charts, through data mining, modeling, natural language processing, and machine learning. Natural language processing is software working with human language. Machine learning is a model that improves as it sees more examples. O*NET says most people in the role need considerable preparation, often a four-year degree.

Copy a wage from a named page on the day you apply, with the date beside the figure. In the 2025 Stack Overflow Developer Survey, 84 percent of respondents used or planned to use AI tools, up from 76 percent, and 51 percent of professional developers used them daily. Python's share of respondents rose 7 percentage points from 2024 to 2025. For kronor, use a live Swedish job ad or an official wage table.

OECD.AI estimates AI skill demand from job ads on Adzuna in 16 countries. Sweden is outside that set. The AI group covers software development, research and methodology, machine learning tools, data management, and web development.

A first role usually asks you to follow a written rule and show finished examples. An evaluation role asks for a test that can fail: fixed cases, a score, and a line on what the score misses.

Three CV lines and five interview questions

I am applying for an AI role. Below are notes from a project I did. Rewrite them as three CV lines. Each line starts with a verb, names what I made, and names how I checked it. Then write five interview questions an employer might ask, each with a one-line hint for a strong answer. Mark which questions fit a first role (follow a rule, show examples) and which fit an evaluation role (a test that can fail, a score, a known miss). Do not add tools, employers, or results that are absent from my notes. My notes: [paste your project notes here]

Expert

Specialist work you can prove

Fine-tuning trains an existing model further on a smaller set of examples for one job. The job ad should name who owns the examples, the score that separates a better model from a worse one, and what happens to the data when the run ends.

Distributed inference answers with a model that is too large or too busy for one chip, by splitting the work across several processors. Ask for the time limit in seconds, the size of one request, and what the service does when a machine drops offline.

An AI architect chooses how models, stores, and checks fit together. Infrastructure keeps that running: machines, queues, logs, and the path from a saved model file to a live service.

Entry means the rule is given and the person applies it. Working means the person changes a workflow and can show errors or time before and after. Specialist means the person owns a measure, a known failure case, and a decision about hardware or data.

A portfolio another person can run needs a public Git repository (Git records each file change) and, for a model or a dataset, a page on Hugging Face, a public site for sharing models and datasets. The README states the question, the data you were allowed to use, the run command, the score, and one case the score gets wrong. Use ESCO for the same skill name in Swedish and English. Employer posts on Polaris will be free to submit and checked by a person before they are public.

A requirements list for a specialist role

Turn the notes below into a one-page requirements list for a specialist AI role. Use these headings: Goal, Data (who owns it, what may be stored), Measure (the score, and one case where the score can look good while the result is wrong), Hardware limit (time per answer, and what happens if a machine drops offline), Level (Entry, Working, or Specialist, and why), Portfolio links to ask for (a Git repository and, if there is a model or dataset, a Hugging Face page). If a heading has no fact in the notes, write missing under it. Do not invent numbers. Notes: [paste your notes here]

Sweden compared with other markets

Question Sweden Wider published sets
Occupation name SSYK 2012. Code 2512 is software and systems developers. Code 3512 is IT support technician. The full hierarchy, with job titles under each code, is CC0 from JobTech. United States: O*NET 15-2051.00, data scientists (most need a four-year degree; some do not). EU labels for the same skills: ESCO, in Swedish and English.
First role Follow a written rule, show finished examples, and say what you checked by hand. The same test fits annotator work and junior support or automation work.
Evaluation role Bring a fixed set of cases, a score, and a sentence on what the score misses. Specialist cards: fine-tuning, distributed inference, architecture, infrastructure.
Skill demand Read current Platsbanken ads through the open Job Search service, and older ads through the historical service. Both are CC0. OECD.AI uses Adzuna ads in 16 countries. Sweden is outside that set. In the 2025 Stack Overflow Developer Survey, 66 percent named almost-right answers as the top frustration, 45 percent said debugging that output takes more time, and 46 percent distrusted AI-tool accuracy while 33 percent trusted it.
Pay Copy the figure from a live Swedish job ad or an official wage table, and write the date beside it. The Stack Overflow survey publishes pay for the people who answered. Open the 2025 results and copy the figure with the year. The sample is self-selected.

How a listing gets onto the board

Employers cannot post their own job ads here yet. That opens later.

When it opens, an employer uses a form: title, tasks, place, a level written in words (Entry, Working, or Specialist), and a link to the application. A person reads the form before it is public. The check looks for a real organization, a task list, and a level that matches those tasks.

Swedish ads may also be taken from Arbetsförmedlingen's open job-ad data, which is Creative Commons CC0. Job Search covers ads on Platsbanken. The historical API covers ads from 2016 onward that have left Platsbanken, with skills and remote-work fields added. A downloadable historical file holds about 6.9 million ads since 2006. Each card keeps the source link and the date it was read.

Polaris will not invent ads. It will not paste product logos taken from a job ad. A card with no source link and no checked employer form stays off the board.

README skeleton

# Project title
Question this answers:
Data I was allowed to use:
Command to run:
Score:
One case the score gets wrong:
Level I am claiming: Entry / Working / Specialist
Git repository:
Hugging Face page (model or dataset), if any:

The shared station could sort the open job ad feeds by family and level overnight. About the shared station

Sources

Back to top