README.md

metadata

language: ja
license: cc-by-4.0
task_categories:
  - question-answering
  - text-generation
tags:
  - data-engineering
  - japanese
  - instruction-tuning
  - sft
  - pretraining
  - synthetic
pretty_name: データエンジニア2026 学習データセット（日本語）
size_categories:
  - 1K<n<10K

Data Engineer 2026 (Japanese) Dataset

A Japanese-language dataset for training and fine-tuning models on modern data engineering topics as of 2026. It contains instruction-tuning (SFT) Q&A pairs and a pretraining corpus covering distributed processing, streaming, data lakehouse, orchestration, transformation, local analytics, modeling, quality and governance, cloud and data platforms, MLOps, and AI/agentic data workflows.

Dataset splits

Split	Size	Description
sft	1,780 samples	Instruction-following Q&A pairs in Japanese (instruction / input / output). Covers definitions, comparisons, how-tos, design and best practices, debugging, interview-style questions, 2026 trends, outlines, and aspect-specific questions.
pretrain	186 samples	Japanese technical explanatory text (chunked articles and notes) about data engineering topics, intended for continued pretraining.

Schema

SFT (sft)

Column	Type	Description
id	str	Sample ID (sft_N)
category	str	High-level category (分散処理エンジン, ストリーミング基盤, DWH・SQL, テーブル格式, オーケストレーション, 変換, ローカル分析, モデリング, データ運用, クラウド基盤, AI・ML, 現場, トレンド, 移行, 設計, ...)
topic	str	Topic slug (e.g. spark, flink, kafka, iceberg, airflow, dbt, duckdb, polars, lakehouse, contracts, governance-ai, ...)
qtype	str	Question type (faq, definition, overview, design, howto, debug, trend, interview, compare, outline, aspect, variant)
aspect	str	Aspect (general, 全体像, 設計, パフォーマンス, コスト, 運用, セキュリティ, 移行, 2026年動向)
difficulty	str	intermediate (default)
instruction	str	Question or instruction in Japanese
input	str	Optional context (empty in this release)
output	str	Answer or explanation in Japanese (markdown)
lang	str	ja
source	str	curated+synthetic or synthetic

Pretrain (pretrain)

Column	Type	Description
id	str	Sample ID (pre_N)
text	str	Explanatory Japanese text (markdown-style article or FAQ block)
topic	str	Topic slug
category	str	Category
lang	str	ja
source	str	synthetic

Coverage (key topics)

• Distributed engines: Apache Spark 4.x, Apache Flink 2.x
• Streaming: Apache Kafka 4.x (KRaft), streaming fundamentals, CDC (Debezium), streaming databases (RisingWave / Materialize)
• Lakehouse and table formats: Apache Iceberg, Delta Lake, Hudi, catalogs (Unity Catalog / Polaris / Gravitino / Nessie), file formats (Parquet / Arrow / Avro)
• Warehouses and SQL engines: BigQuery, Snowflake, Redshift, Trino, StarRocks, ClickHouse, SQL craft
• Orchestration and transformation: Apache Airflow 3.x (Task SDK), Dagster (SDA), dbt (Fusion / Core v2.0), SQLMesh
• Local and modern analytics: DuckDB 1.5 / 2.0, Polars, Python data stack 2026 (pandas 3.x, uv, ruff, marimo, pydantic)
• Data modeling and contracts: Kimball / Data Vault / OBT, data contracts (ODCS), data quality (Soda / Great Expectations / Deequ / Elementary), lineage (OpenLineage)
• Cloud and platform: AWS / GCP / Azure data platforms, Kubernetes data (Spark / Flink operators / Strimzi / Kueue)
• AI / ML and data for AI: AI/LLM data pipelines (embeddings / RAG / vector DB), MLOps data (feature store / lineage / versioning), governance for AI agents (MCP / text2SQL)
• Ops / FinOps / reliability, testing / CI, migration, design patterns

All content reflects 2026-era trends (e.g. Iceberg as the open table format, Spark Connect, Flink disaggregated state, Airflow 3, dbt Fusion, DuckDB / Wasm / Iceberg, streaming-first lakehouse, data contracts, agentic data workflows).

How to load

pip install datasets

from datasets import load_dataset

ds = load_dataset("LoNebula/data-engineer-2026-ja")
print(ds)
# ds["sft"], ds["pretrain"]

Notes and license

• This dataset is primarily synthetically generated from a curated Japanese knowledge base. It is intended for research, educational purposes, and model training or evaluation.
• While care was taken for factual accuracy (aligning to 2026 versions and trends), it may contain minor inaccuracies. Always verify against official documentation.
• License: CC BY 4.0. Retain attribution to LoNebula and a link to this dataset when redistributing or building derived work.

Acknowledgments

Generated with assistance from Kilo / Claude Code tooling. The dataset structure and schema are designed for instruction tuning (SFT) and continued pretraining in Japanese.