Daniel Cárdenas

Full-Stack Builder: Firmware · Web · AI

I design and ship end-to-end systems—from STM32 firmware and edge data capture to Go/Python backends and multilingual RAG experiences.

XPLevel 1
0 XP150 to level up
Jun 2024Firmware → Full-stack Data Tools2 min read

Data Reconciliation Workbench

Left/Right source comparison across DB/Excel/CSV with tolerance rules and Excel audit report.

Excel AutomationSchema MappingTolerance Rules
PythonPyQt6pandasSQLiteExcel

Confidential Context

This case study is sanitized. Client data and proprietary integrations are omitted. Built at an energy market operator (Colombia). This page shows a sanitized demo.

Outcomes

  • Time to reconcile: 4–6h → <30m
  • Mismatch rate reduced ~70% after rule tuning

Problem

An energy market operator reconciled dispatch schedules across SCADA, Excel, and broker uploads. Analysts manually eyeballed differences, creating long delays and inconsistent audit trails during regulatory reviews.

Approach

  • Partnered with operations SMEs to map tolerance rules and exception workflows.
  • Assembled a PyQt6 desktop app with modular plugins that load spreadsheets, SQL queries, or CSV drops.
  • Crafted a diff engine in pandas that normalizes units, applies tolerance bands, and flags anomalies with explanations.
  • Generated Excel audit packs with macros disabled, including pivot-ready tabs and sign-off cells for compliance.
  • Published a sanitized lite demo repo: mocked datasets + walkthrough scripts for recruiters and partners.

Results

  • Reduced reconciliation time from 4–6 hours to under 30 minutes.
  • Cut mismatch rate by ~70% through iterative rule tuning and retention of prior decisions.
  • Enabled non-technical analysts to triage exceptions while leaving an immutable evidence trail.