Ukraine IT labour-market demand, 2024–2026 - preliminary analysis

Sources collected: Djinni, DOU (jobs.dou.ua), LinkedIn, robota.ua, work.ua.

Dimension semantics (never mixed in one table): tech/role = the board's own categories; *_search = keyword-search counts for the same term list on every board; total = the board's published total; *_derived = arithmetic over categories with the formula in note.

DOU (jobs.dou.ua) - DOU vacancies published per month (all categories), sum of monthly published values by year:

work.ua - work.ua vacancies per month (all categories), sum of monthly published values by year:

robota.ua - robota.ua vacancies published per month (all categories), sum of monthly published values by year:

Cross-source check over the window both sources cover (2025-09 → 2026-08). Read this before quoting any single series as "the market":

Djinni's two published series fall far more than DOU's postings in the same window, so the Djinni decline is at least partly Djinni-specific (its series cover subsets - salary-listed vacancies, vacancies that received applications) and must not be read as the whole market shrinking at that rate.

Djinni - board categories, 2026-09-24 (share = % of this table):

DOU (jobs.dou.ua) - board categories, 2026-09-24 (share = % of this table):

DOU (jobs.dou.ua) - technology demand change, 2026 vs 2025 (matched periods only, max 12 shown each way):

Each row: the board's latest value for that exact item, its share of the set, and the year-over-year change where the board publishes enough history (matched months).

Technology keywords - counts on the term set every board carries (20 terms, sources: djinni, dou, linkedin, workua):

Share of that common term set, %%:

Role keywords - counts on the term set every board carries (9 terms, sources: djinni, dou, linkedin, workua):

Share of that common term set, %%:

* = LinkedIn bucket (floored to the nearest thousand), an upper-bound floor, not an exact count. Shares are computed on the common term set only, so a board that lacks a term is not credited with it - raw counts are NOT comparable across boards. Keyword semantics differ per board: DOU's ?search= also matches category names (upper bound); work.ua contributes only terms its own engine filed under its IT sphere; robota.ua has no IT-scoped keyword search and is absent; LinkedIn floors counts at 1,000. Per-row details in note.

Djinni - board categories, 2026-09-24:

DOU (jobs.dou.ua) - board categories, 2026-09-24:

DOU (jobs.dou.ua) - derived aggregates, 2026-09-24:

DOU - experience level, share of that month's postings %:

DOU - location, share of that month's postings %:

DOU (jobs.dou.ua) (tech, 2026 vs 2025):

DOU (jobs.dou.ua) (role, 2026 vs 2025):

1. Category taxonomies overlap - a posting can sit in several categories, so column sums exceed board totals; never present a summed category column as a market total.

2. Two different measures coexist: total/tech/role rows from DOU trends count postings published in a month (a flow); live snapshot rows count vacancies open today. They are not comparable and are kept in separate periods.

3. Coverage differs by source: DOU publishes machine-readable per-category history (2023-09 → 2026-08); Djinni exposes a 12-month per-category window plus an API snapshot; work.ua publishes a total series only (no per-technology history); robota.ua publishes aggregate quarterly/monthly totals; LinkedIn has no Ukraine history at all.

4. Per-framework history (React/Angular/Vue) does not exist publicly on any board's category taxonomy - those are only measurable as current keyword snapshots. Treat framework-level trend claims as unsupported by this dataset.

5. Live snapshots are point-in-time (2026-09-24) and move daily; the workbook records the collection date of every row.

6. LinkedIn buckets any count ≥1000 (floor to thousand), so broad terms are floors rather than counts; windowed (7/30/90-day) rows are exact but measure a different window and live in *_windowed_7d/30d/90d dimensions. LinkedIn also falls back to loose/related matching when a term matches nothing (a nonsense keyword returned '3,000+'), so bucketed values are fuzzy-match floors.

7. DOU ?search= matches category names as well as titles - keyword rows are an upper bound for DOU.

8. General boards are cross-industry. work.ua and robota.ua cover every sector, so their keyword counts are only comparable with the IT boards when the board itself classifies the query under its IT sphere - e.g. work.ua sends 'Java' to the manufacturing sphere (27) and 'Security' to its security-guard category (9,389). Those rows are excluded from shares and listed in data/processed/out_of_scope_rows.csv; the IT-sphere counts (React 25, .NET 26, Python 48, C# 28, Node.js 17 at collection time) are used.

9. Djinni tech_salary_only rows count only vacancies with a published salary range (a biased subset, ~5–8k/month) - useful as a trend shape, not as a market size.