Miguel de Queiroz Fernandes Soares

Created automated data pipelines to extract, process, and organize audio from 3,000+ Brazilian Supreme Court (STF) sessions, producing datasets for Portuguese NLP and Speech-to-Text research. Researched and curated Portuguese AI datasets including Canarim, TTS-Portuguese-Corpus, Portuguese-OCR-Dataset, and YODAS-PT using Python with AWS EC2 and S3.

Built TraceCheck end-to-end as the primary engineer, a digital privacy product that helps 100+ users discover personal data exposure and generate actionable privacy protection plans. Designed and developed core features including personal data research, analysis, and report generation, as well as Tracy, an AI conversational assistant. Implemented production systems and automation such as cron jobs, CI/CD pipelines, and product analytics using PostHog, leveraging Next.js, TypeScript, PostgreSQL, Drizzle, and LLMs.

Developed a multi-agent AI system for automated ENEM essay evaluation in Brazilian public schools, designing specialized agents and correction workflows with Python and OpenAI models, achieving 96% agreement with human graders. Built asynchronous backend pipelines supporting 4M+ essays and 1M+ users, including peaks of 224K essays processed in a single day. Helped the platform earn a Guinness World Record for 461K+ AI-corrected handwritten essays in one month, and contributed to AI-powered lesson planning and presentation generation tools for teachers.