Role
Linguistic curation, product, and implementation
Humanizador PT-BR
An MCP server integrated with ChatGPT that detects and rewrites AI markers in Brazilian Portuguese. It is calibrated against real PT-BR writing patterns, not criteria imported from English.
Project summary
Audience
Creators, content teams, and AI users writing in PT-BR
Tools
MCP, TypeScript, Zod, Claude, Bedrock, AWS
Outcome
Distributable tool to detect and rewrite AI markers
Context
A proprietary tool for reviewing AI-generated text in Brazilian Portuguese.
Business problem
Many people use AI to write, but the output often still reads as machine-generated. In PT-BR this shows up in very specific ways, and generic tools don't catch it properly.
Audience & personas
- Content professionals
- Consultants and creators who use AI for drafts
- Teams that need to maintain brand voice in Portuguese
Adoption barriers
- Highly subjective review criteria
- Confusion between formal text and artificial text
- Few references calibrated for Brazilian Portuguese
Solution design
- Skill with versioned editorial rules
- MCP server with separate tools for detection and rewriting
- Review guided by register, meaning, and intent
Assets created
- Guide to AI patterns in PT-BR
- MCP tool for ChatGPT
- Infrastructure with authentication, rate limiting, logs, and tests
Metrics & signals
- 30 linguistic patterns documented
- 2 usage modes: detection and rewriting
- Serverless deployment on AWS
Lessons learned
The best review doesn't try to 'beautify' the text. It removes the artificial and lets the intent come through.
Next improvements
Create presets by register: corporate, technical, social, editorial, and academic.