Alexandre Rodrigues Fernandes

Engineered a production RAG + Text-to-SQL system end-to-end, from architecture through deployment, optimized from 13% to 77% accuracy on a 30-query domain golden set using a fully local Qwen2.5 7B model. • Built an agentic SQL pipeline with LangGraph (deterministic pipeline + ReAct fallback), BM25+MMR hybrid retrieval, dynamic few-shot prompting, and schema pruning across 500+ tables in 3 PostgreSQL databases. • Designed a domain-specific evaluation framework using LLM-as-Judge.