Import
CSV, TSV, JSON arrays, JSON objects, and objects containing a data array.
SOFTWARE / DATA TOOL
Data Workbench
A browser-based tool for inspecting, searching, editing, analyzing, and benchmarking structured datasets.
23,220
rows tested
7
columns in test dataset
5,000
benchmark queries
3
lookup strategies
01
Overview
Data Workbench began as a Data Structures and Algorithms course project and was later redesigned independently into a more general structured-data exploration tool.
The current application can load real CSV, TSV, and JSON datasets, normalize their structure, search and sort records, calculate numeric summaries, modify data, export CSV files, and compare different lookup strategies.
02
Capabilities
Explore
Search globally or by column, sort fields, and browse records through pagination.
Edit
Add, modify, and delete records directly inside the current dataset.
Analyze
Automatically calculate mean, median, minimum, and maximum for numeric fields.
Benchmark
Compare linear search, a Map index, and sorted-array binary search.
Export
Export the current edited dataset back to CSV.
03
Architecture
File
CSV / TSV / JSON
→
Parse
Schema discovery
→
Normalize
Record model
→
State
Working dataset
→
Operate
Search / Analyze / Edit
Schema handling
Duplicate CSV headers are retained safely, blank headers are ignored, and malformed extra fields do not silently expand the schema.
JSON normalization
JSON schemas are generated from the union of keys found across the loaded records.
Numeric inference
A field is classified as numeric when at least 80% of its populated values parse as numbers.
04
Search Benchmark
The current benchmark measures repeated exact-match lookups while separating preprocessing cost from lookup cost.
23,220 records
value column
5,000 queries
median of 5 runs
| Strategy | Build | Search | Total |
|---|---|---|---|
| Linear Array | 0 ms | 439.1 ms | 439.1 ms |
| Map Index | 1.0 ms | 0.2 ms | 1.1 ms |
| Sorted + Binary | 844.2 ms | 162.4 ms | 1003.3 ms |
Binary search reduced lookup time compared with repeated linear scanning, but constructing the sorted representation introduced a substantial upfront cost.
For this workload, the Map index provided the lowest combined build and lookup time. The experiment demonstrates why data-structure selection depends on the complete workload, not lookup complexity alone.
CURRENT BUILD