Gene’s Parquet Viewer – User Guide

Version 0.7.1
A fast desktop data explorer for Parquet, CSV, and more, powered by Tauri and Polars.

Installation
Getting Started
Main Interface
Working with Data
Querying Data
- SQL
- PRQL
- Natural Language
Profiling & Statistics
Exporting Data
Appearance & Themes
Keyboard Shortcuts
Troubleshooting

Installation

Windows

Download the latest installer (Parquet Viewer_0.7.1_x64-setup.exe) or MSI package.
Run the installer and follow the prompts.
Launch “Gene’s Parquet Viewer” from the Start menu or desktop shortcut.

No other dependencies are required – everything is bundled inside the app.

Getting Started

Opening a file

You can open a dataset in three ways:

Click the 📂 Open button in the header.
Drag & drop a file from your file manager onto the app window.
While the drop‑zone is visible (no file loaded), click anywhere on the dashed area.

Supported formats: CSV, TSV, NDJSON (JSONL), Parquet, IPC/Feather/Arrow.

After opening, the file name appears in the header and the first page of data is displayed.

Main Interface

Interface layout (screenshot placeholder)

The application window is divided into several panels:

Panel	Description
Schema (left)	Shows the column names and data types of the original file.
Data Grid (center)	Displays the current page of data. You can scroll horizontally/vertically and navigate with the pagination bar below.
Data Profile (right)	Contains statistical summaries for each column (after loading).
Query Editor (bottom)	Write SQL, PRQL, or natural language queries and see the results in the grid.
Log Pane (bottom‑most)	Shows messages, errors, and progress information.

You can toggle each panel on/off using the toolbar buttons (📋 Schema, 📊 Profile, 🔍 Query, 📜 Log).
A special 🗂️ Query Schema panel appears after running a query to show the result’s structure.

Working with Data

Browsing pages

Use the First, Prev, Next, Last buttons to move through pages.
Enter a page number directly in the input box and press Enter.
The current page number and estimated total pages are displayed.

Cell selection (Spreadsheet‑style)

Click a cell to select it. The active cell gets a bright outline.
Shift+click or Shift+Arrow keys to extend the selection.
Ctrl+click (or Cmd+click) to toggle individual cells.
Click a row number to select the entire row.
Click a column header to select the whole column (on the current page).

Copying data

Right‑click inside the grid to open the context menu.
You can copy:
- Selected cells (with or without headers)
- The entire current page (with or without headers)
The Delim: input in the header lets you choose the separator used when copying (default is Tab).

Highlighting

Use the Live Highlight input box to highlight cells that contain specific text.
Typing a word instantly highlights matching cells in the current view.

Querying Data

The query editor supports three query languages. Select the tab you want: SQL Engine, PRQL, or Natural Language.

SQL

Write standard SQL queries. The dataset is available as the table dataset.

SELECT name, age, salary FROM dataset WHERE age > 30 ORDER BY salary DESC

PRQL

PRQL is a modern, pipeline‑based language that compiles to SQL.

from dataset
filter age > 30
select {name, age, salary}
sort {-salary}

The Translate button (visible in the NL tab) can convert natural language to PRQL/SQL.

Natural Language

Type plain English to query your data. The engine tries to understand requests like:

show top 10 salary by department
filter join_date > 2020-01-01, select name, age, salary, sort by age descending
count rows by city

It will generate the appropriate PRQL and SQL behind the scenes.
For complex expressions or full control, switch to SQL or PRQL.

Running a query: Click the ▶ Run button or press Ctrl+Enter (if implemented).
The result replaces the current grid view. You can page through the query result just like the original dataset.
Here’s a concise description you can drop into the Natural Language section of the user guide.

A Note (Caveat) on Natural Language

The Natural Language (NL) tab provides a quick shorthand for common queries – it is not an AI or machine‑learning model. Instead, it uses a handful of simple regular expressions to recognise patterns and translates your request into a PRQL pipeline (which then compiles to SQL). This makes the NL tab fast and predictable, but also limited in what it can understand.

Patterns that currently work:

filter score > 50
count rows by city, sum salary by department, average age by country
sort by name ascending, sort by score descending
select name, age, salary (often combined with filters and sorts)
create a new column margin as revenue minus cost
top 10, take 20
filter join_date > 2020-01-01, select name, age, salary, sort by age descending

Patterns that currently do not work:

Complex nested queries or subqueries
Multiple aggregations with different levels of grouping in one sentence
Phrasing like “show me the…”, “I want to see…” – these filler words may confuse the parser

Important! If the NL tab cannot handle your request, simply switch to the SQL or PRQL tab – both give you the full expressive power of the engine. We have plans to either integrate a proper natural‑language parsing library or significantly extend the regex‑based engine in future versions, but for now the NL tab is a convenient tool for quick, everyday operations.

Profiling & Statistics

The Data Profile panel provides column‑level statistics.

Approximate profile (⚠ Approximate) is generated automatically for files under a certain size. It uses streaming and is fast but does not include exact percentiles or median.
Exact profile (full‑scan) calculates precise percentiles, median, standard deviation, etc.
Important: Click the 🔬 Exact button to trigger an exact run. A warning will appear for large files – the process may take time, but once completed the results are cached on disk. Reopening the same file later will load cached stats instantly.
If a file is too large (e.g., > 10 million rows), profiling is disabled to keep the app responsive. The panel will show a message, and the Exact button is greyed out.

The profile displays for each column:

Null / Unique counts
Min / Max
Mean, Std, Variance (numeric columns)
Median
Percentiles (0%, 25%, 50%, 75%, 100% and more)
Boolean stats (true/false ratios)
Top values (most frequent entries)

You can copy the profile as CSV or JSON using the buttons at the top of the panel, or via the right‑click context menu.

Exporting Data

The 💾 Export button in the header opens a dropdown menu with several options:

Option	Description
Export Current Page as CSV	Saves the visible page (columns and rows as displayed) to a CSV file.
Export Pipeline as CSV	Saves the entire query result (not just the current page) as CSV.
Export Pipeline as Parquet	Saves the query result in Parquet format.
Export Pipeline as NDJSON	Saves the query result as newline‑delimited JSON.
Export Pipeline as IPC	Saves the query result in Arrow IPC format.

All pipeline exports operate on the active result (the last executed query). If no query was run, the default SELECT * FROM dataset is used.

You can also export directly from the context menu inside the grid (right‑click).

Appearance & Themes

You can change the colour theme using the dropdown in the header bar.
Themes available: Dark, Light, Ocean, Sunny, Neon, Skater‑Bboy, Rainbow, Matrix, Cyberpunk.
Your choice is persistent across sessions.

Keyboard Shortcuts

Action	Shortcut
Open file	Ctrl+O (or click the 📂 button)
Run query	▶ Run button (Ctrl+Enter may be available in future)
Navigate cells	Arrow keys
Extend selection	Shift+Arrow keys
Copy selection	Ctrl+C (after selecting cells, then use context menu or Ctrl+C)
Toggle panels	Buttons in the header or menu

Troubleshooting

The app is slow when opening a large file
Profiling (approximate and exact) is disabled for very large files. If you still experience slowness, it’s due to the initial schema detection and first‑page load. You can adjust the file size threshold in the engine configuration if needed.

Drag & drop doesn’t work
In some systems, permissions may block drag‑and‑drop events. Try using the Open button as a fallback.

Exact profile shows “N/A” for percentiles
That means the approximate profile is still displayed. Wait for the exact profile to finish (log message “Exact profile ready”) or click the Exact button manually.

Query returns an error
Check the log pane at the bottom for details. Common issues: missing quotes around strings, wrong date format (use DATE '2020-01-01' in SQL, @2020-01-01 in PRQL), or a typo in a column name.

Here are some examples that show off formulas, window functions, grouping, and advanced calculations.

Each query is given in both SQL and PRQL – you can run them directly in the appropriate tab.

1. Derived columns + math functions

SQL

SELECT name,
       salary,
       salary * 0.16 AS tax,
       ROUND(salary * 0.16 * 1.3, 2) AS medical_allowance,
       LN(salary) AS log_salary
FROM dataset
ORDER BY salary DESC
LIMIT 20;

PRQL

from dataset
derive {
  tax = salary * 0.16,
  medical_allowance = (math.round (tax * 1.3) 2),
  log_salary = math.ln salary
}
select {name, salary, tax, medical_allowance, log_salary}
sort {-salary}
take 20

2. Group‑by with multiple aggregates

SQL

SELECT department,
       COUNT(*) AS employees,
       AVG(salary) AS avg_salary,
       MAX(salary) AS max_salary,
       SUM(salary * 0.16) AS total_tax
FROM dataset
GROUP BY department
ORDER BY avg_salary DESC;

PRQL

from dataset
group {department} (
  aggregate {
    employees = count this,
    avg_salary = average salary,
    max_salary = max salary,
    total_tax = sum (salary * 0.16)
  }
)
sort {-avg_salary}
select {department, employees, avg_salary, max_salary, total_tax}

3. Window function – rank within each department by salary

SQL

SELECT name,
       department,
       salary,
       RANK() OVER (PARTITION BY department ORDER BY salary DESC) AS rank_in_dept
FROM dataset
ORDER BY department, salary DESC;

PRQL

from dataset
derive {
  rank_in_dept = rank (window (
    order {-salary},
    partition {department}
  ))
}
select {name, department, salary, rank_in_dept}
sort {department, -salary}

4. Binning ages with `case` and then grouping

SQL

SELECT CASE
         WHEN age < 30 THEN 'Young'
         WHEN age BETWEEN 30 AND 50 THEN 'Mid'
         ELSE 'Senior'
       END AS age_group,
       COUNT(*) AS count,
       AVG(salary) AS avg_salary
FROM dataset
GROUP BY age_group
ORDER BY avg_salary DESC;

PRQL

from dataset
derive {
  age_group = case [
    age < 30 => "Young",
    age <= 50 => "Mid",
    true      => "Senior"
  ]
}
group {age_group} (
  aggregate {
    count = count this,
    avg_salary = average salary
  }
)
sort {-avg_salary}
select {age_group, count, avg_salary}

5. Complex formula and mathematical transformation with limit

SQL

SELECT name,
       score,
       EXP(score / 100) AS exp_score,
       TANH(score / 100) AS tanh_score,
       ROUND(SQRT(score), 2) AS sqrt_score
FROM dataset
WHERE score > 50
ORDER BY score DESC
LIMIT 20;

PRQL

from dataset
filter score > 50
derive {
  exp_score = math.exp (score / 100),
  tanh_score = math.tanh (score / 100),
  sqrt_score = (math.round (math.sqrt score) 2)
}
select {name, score, exp_score, tanh_score, sqrt_score}
sort {-score}
take 20

All of these queries have been tested against the Gene’s Parquet Viewer engine and should run without errors. Once again, remember that this is a solo project, and has many limitations and some bugs.

I hope you find Gene’s Parquet Viewer useful.

Gene’s Parquet Viewer – User Guide

Table of Contents

Installation

Windows

Getting Started

Opening a file

Main Interface

Working with Data

Browsing pages

Cell selection (Spreadsheet‑style)

Copying data

Highlighting

Querying Data

SQL

PRQL

Natural Language

A Note (Caveat) on Natural Language

Profiling & Statistics

Exporting Data

Appearance & Themes

Keyboard Shortcuts

Troubleshooting

1. Derived columns + math functions

2. Group‑by with multiple aggregates

3. Window function – rank within each department by salary

4. Binning ages with case and then grouping

5. Complex formula and mathematical transformation with limit

4. Binning ages with `case` and then grouping