Allan Simões AdegasPortuguês
Menu
Back to home

Data-Matcher

In development

Automatic comparison of two spreadsheets or data files to find, in seconds, what doesn't match.

Role
Author · Asigas product
Started
2026
extrato.xlsx1023410235102361023710239sistema.parquet1023410235102361023810239divergent valueonly in A · only in B
Project illustration
6
formats read
5
kinds of difference reported
API v1
public, with access keys

Overview

Reconciling two datasets (the bank statement and the system, the order and the invoice) is still done with spreadsheet lookups at many companies. Data-Matcher takes one file per side, lets you build the match key visually and returns the differences.

What it does

Any tabular file
Excel, CSV, Parquet, JSON, ODS and Arrow, with format, header, encoding and delimiter detection.
Key built on screen
One or more column pairs, with concatenation, fixed text and transformations (prefix, leading zeros, digits only, number, date), previewed before running.
Actionable results
Only on side A, only on side B, divergent values, repeated keys and blank keys, exportable as CSV, XLSX or Parquet.

Architecture

Architecture: Data-MatcherFile A → Normalization. File B → Normalization. Normalization → Rules. Rules → Engine. Engine → Differences. Engine → API /v1File AXLSX · CSV · …File BParquet · JSON · …Normalizationtext tablesRulesvisual keyEnginePolarsDifferenceson screenAPI /v1integration
From raw file to difference report.

Decisions and trade-offs

Polars for the engine
Rules compile into Polars expressions, which keeps matching fast on large files without Python loops.
A queue in the database
Runs use a queue in PostgreSQL: one fewer dependency to operate, enough for the product's volume.

Quality and operations

  • Unit tests without Docker and integration tests with PostgreSQL.
  • Every format reader, the engine, rule preview, queue, export, sign-in and public API implemented and tested.

Status and next steps

  • The production image was prepared for Pier and validated locally. Publication, the team screen, rate limiting and a run-finished webhook are pending.

Stack

PythonPolarsFastAPIPostgreSQLCloud StorageReactTanStack