# How I Built a 100% AI-Generated Document Translator with WCAG AAA & Zero-Data Retention

> Source: <https://dev.to/luizgarcg/how-i-built-a-100-ai-generated-document-translator-with-wcag-aaa-zero-data-retention-422e>
> Published: 2026-08-28 12:09:21+00:00

Overview:A full-stack, production-grade web application created entirely via AI/Vibe Coding. It converts complex legal terms, fine print, and bureaucratic bills into everyday Portuguese while enforcing WCAG AAA accessibility and strict LGPD compliance.

Recently, I set out to test the limits of **Vibe Coding** by building an end-to-end software product using AI-driven development.

The result is the **Real-World Document Translator** (*Tradutor de Documentos da Vida Real*). Its goal is to bridge the accessibility gap for elderly individuals, people with low vision, and folks with limited literacy—allowing anyone to understand legal and financial contracts without needing a lawyer.

Translating legal documents through AI presents two major technical hurdles:

| Challenge | Impact / Risk |
|---|---|
Privacy & LGPD |
Raw documents contain sensitive PII (CPFs, bank accounts, emails). Standard regex fails due to OCR noise (broken numbers, homoglyphs). |
Accessibility |
Most web tools ignore low-vision users and the elderly, lacking high contrast, scalable typography, and native screen reader tools. |

Through iterative prompting and AI engineering, we designed a zero-trust, highly accessible architecture.

`.env`

variables.[ User ] ───▶️ [ Accessible Frontend (HTML/CSS/JS) ]

│

▼

[ Secure BFF (FastAPI) ]

├── File Validation (Magic Bytes)

├── PII Sanitization (OCR-Resilient)

└── API Key Guard (.env)

│

▼

[ Google Gemini API / Mock AI ]

Building this project 100% through AI was a major technical milestone. It proved that Vibe Coding isn't just about speeding up syntax—it's about directing AI to construct **secure, accessible, and architecturally sound systems**.

**Key takeaways from this build:**

Want to test it locally? Follow these simple steps using Python 3.10+:

```
bash
# 1. Clone repo & install dependencies
pip install -r backend/requirements.txt

# 2. Run the application
python backend/server.py

# 3. Open [http://127.0.0.1:8000](http://127.0.0.1:8000) in your browser
Note: If no Gemini API key is configured, the server automatically defaults to Smart Mock AI mode for instant offline testing!
```