API responses and exports arrive as JSON, which is convenient to produce and expensive to analyse. This converter flattens the nested structure into columns, assigns each one a real type, and writes a compressed Parquet file you can query directly.
Reading JSON
Auto-detect reads the first rows and picks the separator that splits them consistently.
Turn off when the file starts straight into data. Columns are then named column_1, column_2, and so on.
Turns "42" into a number and "true" into a boolean. Values with leading zeros stay text so IDs and zip codes survive.
Which worksheet to convert. Filled in once the workbook is read.
Dotted path to the array of records, such as data.items. Leave empty to use the whole document.
Turns {"user":{"id":7}} into a user.id column. Arrays are kept as JSON text.
Pick the encoding the file was written in. Wrong encoding shows up as garbled accents.
Comma-separated words that should become null instead of text.
Writing Parquet
Snappy is the default across Spark, DuckDB and pandas. Gzip is smaller but slower to read.
Turns a user.id column back into {"user":{"id":7}}.
Excel needs a byte order mark to open UTF-8 files with accents correctly.
Rows per row group. Larger groups compress better; smaller groups let readers skip more.
conversion-preroll
This short ad is what keeps the converter free.
Parquet is unreadable in a text editor. That is by design.Check your new Parquet file with ParquetReader — opens Parquet files directly.Nested, self-describing records that every language can read. Typed, compressed, columnar storage built for analytical scans.
| JSON | Parquet | |
|---|---|---|
| Type information | Strings, numbers, booleans, null | Full — ints, floats, booleans, timestamps, decimals |
| Structure | Nested objects and arrays | Columnar, with nested type support |
| Compression | None | Built in (Snappy, Gzip, Zstd) |
| Typical file size | Largest — keys repeat on every record | Typically 5–15% of the CSV |
| Schema | Implicit, per record | Stored in the file footer |
| Human-readable | Yes, though verbose | No — binary |
| Tool support | Every language, natively | Spark, DuckDB, pandas, BigQuery, Athena |
A tags array becomes one column containing ["a","b"] as a string. Exploding arrays into tags.0 and tags.1 would make your column count depend on the longest array in the file, which turns one unusual record into hundreds of empty columns.
Because columns are the union of all keys, a handful of records with extra fields adds columns that are null everywhere else. Use the record path option to target a consistent subset if this gets noisy.
If id is a number in most records and a string in a few, the column is written as text to avoid losing data. Fixing the inconsistency upstream produces a much better Parquet file.
A directory of JSON responses is awkward to analyse. One Parquet file loads into DuckDB or Spark and answers questions in milliseconds.
Most lake table formats — Delta, Iceberg, Hudi — store their data as Parquet underneath. Converting at the edge saves a processing step later.
JSON repeats every key on every record. Parquet stores each column once, which is why the size difference on large exports is dramatic.
sticky-bottomYour file is downloading
post-conversion