Automated reporting platform

Where it started

Maxcess International brings to Mexico the leading technologies and ecosystems in the security sector for retail: inVue, CrossPoint and Indyme. Their systems protect display cases, control stockroom access and measure store operations in large retail chains nationwide.

Every device leaves a trail: openings, accesses, alerts and response times. That trail is worth something when it arrives on time and in each client's format, and every Strategic Account Manager runs their accounts on their own calendar: weekly, every-10-day, semimonthly and monthly reports. They are read by strategic staff with executive visibility and by operational roles in each chain.

The challenge

The data lived in each brand's platform and came out in different exports per store and per chain: CSV, JSON, HTML and TXT. Downloading, cleaning and assembling each report by hand took hours, and a single wrong figure cost the whole document its credibility.

The challenge wasn't one report but 110+ a month, with 10+ client variants and 10+ calculation variants, without losing data integrity or each account's identity.

How I approached it

Before any code, every metric starts in a glossary: what it measures, where it comes from and how it's validated. The survey with each Strategic Account Manager, the design and the interactivity of each format are written up as specifications before they become SQL flows or screens.

I built it in layers, locally and with tests from day one. I direct AI agents with defined roles—architecture, frontend, optimization, design and quality control—and review every delivery against its specification before merging it.

End to end

Extraction. Python scripts pull the platforms' exports on their own—work that used to be manual and took hours. Inside the platform, the Data Funnel receives each file, detects which account and period it belongs to, removes duplicates and emits a canonical dataset with its quality report.

Model. A canonical contract fixes the columns every batch must carry. The SQL engine runs in the browser with DuckDB and persists in Supabase, on PostgreSQL. Each batch keeps its fingerprint and the engine version that computed it, so any figure can be traced back to its source file.

Output. Report Studio walks each delivery through six steps—setup, data load, metrics, content, preview and export—and produces three formats: the monthly executive PPTX, the detailed analytical report (RAD) in Word and PDF, and an interactive HTML dashboard published every month with QR access and a secure link.

What I built

A multi-user platform with four roles—admin, Strategic Account Manager, support and visitor—where each account is an isolated environment: every Strategic Account Manager sees only their own clients, and the admin audits users and activity.

• Command Center: the state of the cycle on one screen, with an assistant that answers about what your profile can see.
• Data Funnel: loading, detection, merging and validation of each batch, with its quality report and sign-off.
• Report Studio: the six-step wizard that assembles, reviews and exports each report.
• Interactive analytics dashboard: summary, impact events, padlocks, display cases, zones, activity by day and hour, and annexes, with filters and CSV, PPTX and RAD downloads.
• Status by branch: days with data, availability and event signals for each batch.
• Surveys: a scaled editor to place display cases and devices on each store's floor plan, with heat maps by device.
• Log, Task Manager and Cycle Calendar: findings, decisions, to-dos and each month's closes, connected to Microsoft 365.
• Data Repository and Explorer, Dashboard Studio, Formats and Glossaries, Reference Center, My Progress, Automations and Admin.

How I know a figure is right

Before it ships, every report goes through automatic checks: every figure in the text must exist in the data package, every name must be in the catalogs, and no alert can be left unresolved. AI-assisted writing only suggests; if it fails those checks, the report falls back to the deterministic template and says so.

Behind it are 4,000+ automated tests that run on every change, with Vitest, Testing Library and Playwright.

What changed

110+ validated, on-time reports a month. The extraction that took hours now runs on its own, and every account gets its format with its brand, typography and charts.

Clients liked the formats, and they opened new uses: operations, marketing and other teams in the chains now read them too. The HTML dashboard is the first step toward the quarterly and annual comparison.

What I learned

That integrity is designed: a glossary and a contract before the code save weeks of fixes. And that automating well doesn't mean hiding the process—it means showing where every figure comes from.

What's next

Bring the quarterly and annual comparison to the dashboard, integrate the Power BI architecture that runs in parallel and publish the repository when it's ready.