PTEN
Back to portfolio
Proprietary product, language and AI

Humanizador PT-BR

A tool that detects markers of AI-generated writing and proposes a PT-BR rewrite that preserves meaning, intent, register, and voice.

Problem, decision, and evidence

Problem

Portuguese-language editors encountered recurring AI markers, while most available references had been calibrated for English.

Role and decision

I turned editorial criteria into versioned rules and separated detection from rewriting so users can analyze text without being forced to change it.

Available evidence

The development environment demonstrates both modes and the MCP integration. No public benchmark for accuracy or reader preference is available.

Development demo: the detection tool identifies linguistic patterns without rewriting the text.
Development demo: the rewrite mode preserves meaning, register, and intent while reducing artificial markers.

Click to enlarge

Scope of work

Role

Linguistic curation, product, and implementation

Audience

Creators, content teams, and AI users writing in PT-BR

Methods & technology

MCP, TypeScript, Zod, Claude via Bedrock, and AWS

Outcome

Users receive a review that reduces artificial markers without replacing meaning or voice with a generic style

Context

A proprietary tool for reviewing AI-assisted writing in Brazilian Portuguese. The screens on this page show a development environment, not a client deployment.

Business problem

AI-assisted text can keep the right content and still lose the writer's voice. In PT-BR, this appears through specific patterns that criteria translated from English do not handle well.

Audience & personas

  • Content professionals
  • Consultants and creators who use AI for drafts
  • Teams that need to maintain brand voice in Portuguese

Adoption barriers

  • Subjective review criteria
  • Confusion between formality and artificial writing
  • Few references calibrated for Brazilian Portuguese
  • Concern about sending sensitive content for review

Solution design

  • Versioned editorial rules for PT-BR
  • Separate tools for detection and rewriting
  • Review guided by meaning, voice, register, and intent
  • Authentication, request limiting, and log privacy

Scope and deliverables

  • Guide to AI writing patterns in PT-BR
  • MCP server for ChatGPT integration
  • Serverless infrastructure with access controls and observability
  • Automated tests for contracts and flows

Evidence and signals

  • Two demonstrable modes: detection and rewriting
  • Development environment available for a private demonstration
  • No public benchmark for accuracy, preference, or voice preservation

Lessons learned

A useful review removes what sounds artificial without replacing the author's choices with a standardized voice.

Next improvements

Validate presets by register and build an evaluation protocol with documented texts, reviewers, and criteria.

View implementation details
  • The MCP server exposes separate tools for detection and rewriting, with contracts validated by Zod.
  • The existing implementation includes authentication, rate limiting, minimized content in logs, and automated tests.
  • Processing uses Claude through Amazon Bedrock, with Lambda, API Gateway, DynamoDB, and CloudWatch. These controls describe the architecture, not a linguistic quality benchmark.

Related services

AI Prototypes & DemosMVPs, automations, and demos to test an idea's value before scaling — using tools like ChatGPT, OpenAI API, n8n, TypeScript, and AWS to prove what works with a tight scope.

Next cases

Enterprise TrainingFacilitation experience transferable to new topics, including AI programs designed for each contextLapidadorUsers move from a rough idea to a structured prompt with an explanation of the choices
Private demonstration

Want to see how the review works with a text from your context?

The demo can take place in a private conversation using a non-sensitive example, with room to discuss voice, privacy, and evaluation criteria.

Book a private demo