João Victor Pessoa Rocha

Computational linguist and senior prompt engineer with 5+ years of experience designing, testing, and optimizing prompts and datasets for large language models and conversational AI systems. Specialized in multilingual NLP (Brazilian Portuguese, English, Spanish), AI agent development, and data annotation for high-accuracy classification and generation tasks. Solid knowledge of prompt engineering methodologies (PALI and related modular frameworks) applied across enterprise conversational AI deployments for major LATAM brands.

Extensive experience in dataset design, annotation, and quality control, including annotation of 40,000+ texts for classification systems and compilation of 100+ aligned bilingual sentence pairs for machine translation training.

Experience designing and deploying 30+ production-grade prompts across text classification, generation, and conversational use cases, contributing to measurable reductions in escalation rates and improved response accuracy.

Background in sociolinguistic profiling and forensic linguistics through computational methods (Master's research, UFMG). Open to fully remote roles in AI, NLP, and localization.

Senior Prompt Engineer - Rubicon - Remote

(2025-12)

• Led prompt engineering initiatives for a data-analytics AI chat product, designing modular prompt architectures to improve response accuracy and reduce escalation rates.
• Created and managed prompts for the company's Analytics AI agent, improving response quality for data-related queries; defined product requirements for a new feature.
• Conducted quality control across the platform and AI chat, identifying and resolving issues to maintain accuracy and consistency.
• Developed evaluation frameworks for prompt performance, tracking consistency metrics across iterations.

Mid-level Prompt Engineer - Mutant - Remote

(2025-04 - 2025-08)

• Designed modular prompt structures for voice AI agents using the PALI method across 4 major enterprise clients (C&A, Credsystem, Pernambucanas, Itau), serving a combined customer base of tens of millions.
• Implemented conditional logic to improve response accuracy, reducing escalation rates and enhancing user satisfaction.
• Authored training materials and onboarding documentation for prompt engineers, accelerating ramp-up time for new team members.

Conversational AI Analyst - Winx - Remote

(2024-02 - 2025-04)

• Designed prompts for text-related tasks across the client analytics dashboard, improving AI-generated content quality and client-reported satisfaction.
• Analyzed user interactions with AI-generated content using NLP-driven insights to inform product and prompt improvements.
• Localized platform dashboard content into Brazilian Portuguese, English, and Spanish; produced quarterly client performance reports used in client-facing reviews.

Computational Linguist - Data Stand - Remote

(2024-02 - 2024-04)

• Designed and documented datasets ensuring compatibility with machine learning training pipelines.
• Tested and provided structured feedback on a proprietary dataset generation platform, informing product refinements for NLP use cases.

Brazilian Portuguese Linguist / Language Consultant - PTW - Remote

(2023-04 - 2024-02)

• Transcribed Brazilian Portuguese speech data per phonemic transcription guidelines, supporting speech recognition model training.
• Conducted quality control across audio files and transcriptions, ensuring consistency and accuracy for downstream ML performance.

Computational Linguist - Alana AI - Remote

(2022-10 - 2024-01)

• Led a language variation research project end-to-end, from scope definition through team management to delivery.
• Generated 30+ prompts for text-related tasks, optimizing data collection and model training workflows for NLP applications.
• Annotated 40,000+ texts for high-accuracy classification models; established language data management guidelines adopted org-wide.

Data Annotator-Compiler - Laboratory for Experimentation in Translation (UFMG) - Belo Horizonte, MG, Brazil

(2021-06 - 2022-02)

• Annotated 80+ consumer-generated questions for semantic and grammatical function, supporting chatbot development for a Brazilian public institution.
• Compiled 20+ medical surveys and aligned 100+ bilingual (EN-PT-BR) sentence pairs, supporting training and evaluation of a machine translation model.

Postgraduate Degree - Data Science - Descomplica Faculdade Digital (2026-01 - 2026-12)

Master's Degree - Linguistics - Universidade Federal de Minas Gerais (2022-03 - 2024-02)

B.A. - Language and Literature (Linguistics and English) - Universidade Federal de Minas Gerais (2017-01 - 2022-02)