Voca AI — Voice to Prompt
Transforms audio into LLM-optimized prompts — transcription + GPT optimization in a single request.
The Problem
Speaking an idea aloud is fast, but turning it into a structured, effective prompt for LLMs is tedious — most people lose context, repeat speech fillers and don't know how to structure requests to get the most from models.
The Solution
Single-request pipeline: transcription with verbose_json output (includes duration), server-side optimization with graceful degradation if optimization fails, persistence via Prisma + Neon, auth with Auth.js v5 (Google OAuth + credentials bcrypt 12), MIME + magic-byte validation server-side, full OWASP headers (CSP, HSTS, XFO) and 90+ unit tests.
Result
Production app with complete transcription → optimization → persistence pipeline, paginated history, .md/.json export, custom design system (Deep Forest · Soft Canvas · Electric Mint) and LGPD-compliant.
Related Projects
MeteoRAG
Intelligent weather assistant with RAG combining real-time INMET API data with LLMs for natural language queries.
JOX — ETL with LLM
Intelligent ETL pipeline for structured data extraction from agricultural bulletins in PDF using LLM with structured output.
Melisso AI Agent
AI conversational assistant embedded in the portfolio — Claude Haiku streaming, Redis rate limiting and animated interface.