madviking/technology-taxonomy

Queast Technology Taxonomy

A reusable reference dataset for identifying, indexing and organizing software technologies. Maintained by Queast, it connects technology names to stable IDs, classifications, product-family relationships and available descriptive information.

Release 2026.10.7: 1,467 technologies, 117 categories, 16 domains and 1,265 parent relationships. Dataset schema: 1.2.1. License: CC-BY-SA-4.0.

Browse the catalog · JSON API · Top 100 technologies · Original release

Where it is useful

Indexing job advertisements

Use technology names and available aliases to generate candidate matches in job titles and descriptions, then resolve the intended technology from the surrounding text. Store its canonical ID alongside the original mention and source document. This supports consistent search filters, aggregation by technology or category, and comparison of hiring demand across your own document collection.

Keep requirements, optional skills and descriptions of an employer's stack distinguishable. Deduplicate reposted ads before counting demand. A job advertisement mentioning a technology is evidence of a mention or hiring requirement, not proof that the employer has deployed it.

Normalizing CVs and professional profiles

Map technology mentions in CVs or profiles to the same IDs used in job advertisements. This can support searchable technology histories, consistent labels across spelling variants and comparisons between stated experience and role requirements.

Retain the original wording, dates and context in your own application. The taxonomy identifies technologies; it does not measure proficiency, verify experience or supply candidate suitability scores. This dataset contains no CVs, applicant profiles or job-ad texts.

Understanding technology usage and demand

Group technology mentions from documents you are permitted to analyze, such as job advertisements, technical documentation or software inventories. Stable IDs let you compare sources without relying on name strings alone. Categories and documented parent relationships help organize results into useful groups.

Interpret the source correctly: an inventory entry, a documentation mention and a job-ad requirement are different observations. Define your unit of analysis and avoid double-counting a technology and its parent family when aggregating results. The taxonomy itself is reference data, not a measurement of worldwide usage.

Search, entity linking and agent workflows

Use the taxonomy as a lookup vocabulary for search indexing, entity linking, technology directories, knowledge graphs or retrieval-augmented applications. An extraction model can propose a technology name; your application can resolve it to a known identity and retain ambiguous or missing matches for review. The dataset is not an annotated mention corpus, trained extraction model or accuracy benchmark.

Top 100 by job-ad popularity

We also publish a Top 100 technology ranking based on Queast's observed job-ad demand. It combines Finland, Sweden, Germany and the United Kingdom, from January 1, 2026 through the end date shown on the live page.

The ranking orders active leaf technologies by distinct, non-hidden job advertisements linked to them in Queast's index. Technologies with children in the published hierarchy are excluded. The page displays the period and freshness; ranking results are cached for 24 hours. Absolute match counts are not published in the ranking.

This measures job-ad demand within those four observed markets. It does not represent all of Europe, installed software, overall adoption or market share. Ranking JSON and methodology are available separately.

The live Top 100 and weekly job-ad graphs are not included in this immutable release. Some technology records contain a frozen popularity tier (Very rare, Rare, Common or Very common); missing tiers are unknown. These labels are not numerical ranks and do not automatically update when the live ranking changes.

What is included

File	Contents
taxonomy.json	Exact original full release, including available descriptions, evidence, official links and richer record metadata.
taxonomy-only.json	Exact original compact identity-and-structure export, without descriptive editorial prose.
technologies.jsonl	Convenient index view, one technology per row; default Hugging Face subset.
categories.jsonl	All 117 category records.
domains.jsonl	All 16 domain records.
relationships.jsonl	All 1,265 typed edges; this release contains parent_of relationships.
SCHEMA.md / INDEX-SCHEMA.md	Original release schema and the derived table definitions.
CHANGELOG.md / UPSTREAM-README.md	Original release notes and download instructions.
PROVENANCE.json	Exact source commit, release metadata, counts and transformation details.
SHA256SUMS	Checksums for the files in this package.
LICENSE / CITATION.cff	License text and citation metadata.
example.py / build_tables.py	Local usage example and deterministic table conversion.

The two original JSON exports and their .sha256 files are preserved byte-for-byte. JSONL files are convenience views derived from that same release, not a separately edited taxonomy. The default technology table contains selected fields; use taxonomy.json for the full available metadata. train is the Hugging Face loading split name for each reference table, not a supervised training/evaluation split.

Quick start

Load a table with the Hugging Face datasets library:

from datasets import load_dataset

technologies = load_dataset(
    "madviking/technology-taxonomy", "technologies", split="train"
)
categories = load_dataset(
    "madviking/technology-taxonomy", "categories", split="train"
)
relationships = load_dataset(
    "madviking/technology-taxonomy", "relationships", split="train"
)

print(technologies.num_rows)  # 1467 in release 2026.10.7
print(technologies[0]["canonical_id"], technologies[0]["name"])

For reproducible work, add revision="<Hugging Face commit SHA>" to each call after choosing a published repository revision. main can change when a later release is uploaded.

To work with the full release without additional dependencies, download taxonomy.json and taxonomy.json.sha256, then:

shasum -a 256 -c taxonomy.json.sha256
python3 example.py taxonomy.json PostgreSQL

example.py verifies the sibling checksum, lists exact case-insensitive name/alias candidates and shows their canonical IDs and categories. It is a lookup example, not a production entity extractor. Never silently pick the first candidate when a mention is ambiguous.

Identity, hierarchy and matching

• Use canonical_id as identity. Names and slugs are labels or navigation handles; external-system IDs belong to their own namespaces.
• Aliases are incomplete. Only 208 of the 1,467 records in this release have at least one released alias. No match does not establish that a technology does not exist.
• requires_context is explicit for every technology; 194 records are flagged in this release. A true value calls for contextual disambiguation. False does not guarantee that every string occurrence is a technology mention.
• Category IDs join to categories[].id; domain IDs join to domains[].id. Categories are browse classifications. Technology parent relationships represent product families, components or documented ecosystems.
• A parent_of edge points from source_id (parent) to target_id (child). The technology table's parent_canonical_id, category names and domain names are derived conveniences.
• Preserve punctuation when matching names such as C++ or .NET, and retain the source mention and context alongside resolved IDs. Parent membership does not imply equivalence between products.

Provenance, scope and limitations

Queast maintains the taxonomy for internal use and public reuse. This package mirrors the published 2026.10.7 release, generated at 2026-10-05T05:09:03.917203Z. Its original full JSON SHA-256 is 17ad80911fe1f35197be8da05b97199c698aae7488cfb7fb790492930024e77d.

Coverage and optional record details remain incomplete. Public sources, review timestamps and field evidence are included where available; check them before relying on a particular claim. Missing fields mean missing information, not a verified negative. A record's lifecycle value does not guarantee current vendor support.

The taxonomy reflects Queast's editorial scope and technology-discovery process, including European IT hiring evidence for new additions. It should not be treated as an exhaustive or geographically neutral inventory of all technologies. See the public methodology. No raw job advertisements, CVs, private review records or application credentials are included.

The public website and API serve their currently selected release. Check their release version when combining live responses with this pinned snapshot. Future Hugging Face uploads are separate publication events; this repository does not promise automatic synchronization.

License and attribution

This release is distributed under Creative Commons Attribution-ShareAlike 4.0 International. See LICENSE for the terms. The dataset license does not license the referenced software products, third-party trademarks or Queast's application code.

Suggested attribution:

Queast technology taxonomy, release 2026.10.7, maintained by Queast. CC-BY-SA-4.0. https://technologies.quea.st/

The JSONL tables are derived views of the original release; their transformation is documented in INDEX-SCHEMA.md and reproducible with build_tables.py. Identify your own changes when redistributing adaptations and follow the license terms.

For corrections, consult the current contribution status. Hugging Face discussions can be used to report packaging or loading problems; taxonomy changes still require editorial review.