Open data on AI training jobs
The numbers on TrainerGigs are free to download as 5 CSV and JSON files, last built on . No key and no sign-up. Fetch a file with a plain GET, use it, and credit the source.

The files
Every address is on https://trainergigs.com.
| Address | Format | Size | Updates | What it holds |
|---|---|---|---|---|
/ai-training-jobs-roles.json | JSON | 1 to 5 MB | With each site build, about once an hour | Every open listing, one line each: 3,292 on Nov 18, 2026. That is 2,841 roles and 451 paid research studies. |
/ai-training-jobs-data.json | JSON | Under 100 KB | With each site build, about once an hour | Open roles by platform and field in 157 rows, with the day's totals and the counting method. |
/ai-training-jobs-data.csv | CSV | Under 10 KB | With each site build, about once an hour | The same 157 rows as the daily JSON, one row per platform and field. |
/ai-training-jobs-weekly.json | JSON | Under 10 KB | After each week closes | Every closed week, 6 so far, each with its totals. The counting method is inside. |
/ai-training-jobs-weekly.csv | CSV | Under 1 MB | After each week closes | Every closed week as one long table of 1,044 rows, by platform and field. |
To see a file's day, read date in the daily file, counted in the roles file and closed_on on a week.
Every open role: ai-training-jobs-roles.json
Top level: source (the site's address), counted (the day the roles were counted), license (the license and the credit line), role_count_rule (the role-counting method, with a stable id and its definition), comparability (when these counts can be compared with other datasets), fields (what each role field holds, in plain words), totals (the same numbers as totals in the daily file), roles (the listings, sorted by platform, then title, then address).
Each item in roles has these fields.
| Field | Meaning |
|---|---|
id | The role's key on this site: the last part of its address, the same for as long as the role is open. |
url | The role's own page on trainergigs.com, with the tag src=api. Link to it as it is. |
title | The listing's title. |
platform | The platform that lists the role. |
field | The board's name for the role's field of work. |
kind | role, or study for a paid research study. Studies are counted apart and are never called roles. |
pay_min | The lower end of the listed pay as a number. null when the listing states no pay, and when it states a top only ("Up to $90/hr"). |
pay_max | The upper end of the listed pay as a number. null when the listing states no pay, and when it states a bottom only. One rate is both ends. |
pay_unit | The unit of the listed pay: hr, task or yr, or null when the listing states no pay. |
paid_per_task | true when the listing gives an hourly figure but says it pays per task, otherwise false. |
where | The board's short place text, empty when no country is named. |
countries | The two-letter codes of the countries the listing names for applicants, in its order. Empty when it names none. |
posted | The day the platform itself dated the listing, or null when it gives none. |
The rows marked role add up to totals.open_roles. The rows marked study add up to totals.paid_research_studies. The file holds no apply address and no description. Link to a role through its url.
The day's counts: ai-training-jobs-data.json and ai-training-jobs-data.csv
Top level: name (the file's name in words, with its day), date (the day the numbers were counted), source (the site's address), method (how the numbers are counted, one sentence per item), role_count_rule (the role-counting method, with a stable id and its definition), comparability (when these counts can be compared with other datasets), totals (the day's numbers as the site's pages state them), fields (the names of the fields of a row, in CSV column order), rows (one row per platform and field).
The fields of totals
| Field | Meaning |
|---|---|
open_roles | Sum of open_roles over the rows where kind is role. |
platforms_with_open_roles | Platforms with at least one open role. |
paid_research_studies | Sum of open_roles over the rows where kind is study. |
roles_that_state_pay | Roles that state pay in any unit. |
roles_with_hourly_pay | Roles that state hourly pay. |
median_listed_hourly_usd | Median listed hourly pay in US dollars. |
q1_listed_hourly_usd | First quartile of listed hourly pay. |
q3_listed_hourly_usd | Third quartile of listed hourly pay. |
The fields of a row
The CSV has the same columns, in the same order.
| Field | Meaning |
|---|---|
platform | The platform's name. |
field | The board's name for the field of work. |
open_roles | Listings open on the day. |
roles_with_hourly_pay | Of those, listings that state hourly pay. |
median_listed_hourly_usd | Median listed hourly pay, empty when fewer than 8 roles state hourly pay. |
posted_last_7_days | Roles the platform itself dates within the last 7 days. |
countries_named | Countries the listings name, or the platform's published list. |
feed_read | The day the platform's listings were read. |
kind | role or study. |
Every closed week: ai-training-jobs-weekly.json and ai-training-jobs-weekly.csv
Top level: name (the file's name in words), source (the weekly index's address), method (how a week is counted, one sentence per item), notes (how to read the fields of a week), role_count_rule (the role-counting method, with a stable id and its definition), comparability (when these counts can be compared with other datasets), revisions (the weeks that were counted again, and why), weeks (one item per closed week).
- A week runs Monday to Sunday, UTC.
last_dayis its day of record. open_sunday,open_prev_sunday,new_rolesandgone_rolescount every listing, paid research studies among them.roles_sunday,roles_prev_sunday,roles_newandroles_goneare the same counts with the research studies left out.renewed_newandrenewed_gonecount ids that are the same role as one gone or new. They are neither new nor gone.roles_100_pluscounts roles whose listed hourly pay reaches $100.urlis the week's own page on this site.
Each item in weeks has these fields: week, week_start, week_end, last_day, days_counted, days_missing, closed_on, revised, open_sunday, open_prev_sunday, studies_sunday, listed_any_day, new_roles, renewed_new, gone_roles, renewed_gone, newly_tracked, old_postings, roles_100_plus, roles_100_plus_prev, roles_with_hourly_pay, median_listed_hourly_usd, q1_listed_hourly_usd, q3_listed_hourly_usd, platforms_read, feeds_not_read, languages_named, countries_named, days_listed_median, days_listed_n, roles_sunday, roles_prev_sunday, roles_new, roles_gone, platforms_with_roles, url.
The CSV has one row per week, platform and field. Its columns are: week, week_start, week_end, last_day, days_counted, closed_on, revised, platform, field, open_sunday, open_prev_sunday, listed_any_day, new_roles, gone_roles, renewed, newly_tracked, old_postings, roles_100_plus, roles_with_hourly_pay, median_listed_hourly_usd, q1_listed_hourly_usd, q3_listed_hourly_usd, platforms_read, feeds_not_read, languages_named, countries_named, days_listed_median, days_listed_n.
Platform and field are blank on the week's total row. Field is blank on a platform's row. Platform is blank on a field's row. The week's own numbers, from platforms_read to days_listed_n, are on its total row only.
Each closed week also has its own page and files, linked from the weekly index.
Examples
Curl
curl -s --compressed https://trainergigs.com/ai-training-jobs-roles.json -o ai-training-jobs-roles.jsonPython, standard library only
import json
import urllib.request
ask = urllib.request.Request("https://trainergigs.com/ai-training-jobs-data.json", headers={"User-Agent": "my-project/1.0"})
with urllib.request.urlopen(ask) as answer:
data = json.load(answer)
print(data["date"], data["totals"]["open_roles"])JavaScript, in a browser page or in Node 18 or newer
const answer = await fetch("https://trainergigs.com/ai-training-jobs-roles.json")
const data = await answer.json()
const found = data.roles.filter(r => r.kind === "role" && r.countries.includes("NG"))
console.log(data.counted, found.length)How the files are served
- No key. There is nothing to sign up for and nothing to send but a GET.
- Name your program in the
User-Agentheader, as the Python example does. - A file may be kept for 5 minutes, so asking more often than that returns the same file.
- The files send
Access-Control-Allow-Origin: *, so a web page on any site can read them withfetch. - A field may be added to a file at any time. Read the fields you know and pass over the rest.
API description
Every address and every field is described for tools that read OpenAPI 3.1.0: /openapi.yaml and /openapi.json. The two are written from one source. The API catalog is at /.well-known/api-catalog.
Embed a jobs widget on your site. Choose a field or country and copy the code.
License and credit
The data files are licensed under the Creative Commons Attribution 4.0 International license (CC BY 4.0). You may copy, share, adapt and use the data for any purpose, including commercial use, if you give credit. Use this credit line and link:
Data: TrainerGigs, https://trainergigs.com
Say which day the numbers were counted. Say if you changed them.
Role titles and descriptions belong to the platforms that list them. Pay is as each listing states it.
Method and contact
The rules for every count are on the method page, and inside the JSON files as method.
A correction, or a role or platform to add: [email protected]
New to this work? Lesson 1 of our free school takes about 5 minutes, and you can get a badge.
New this week · Entry-level jobs · Jobs by country · Highest pay · Weekly index · All guides · Every role on the board · Every platform · How we count