Bernardo Leite

I’m a final-year Data Science & AI student who builds practical LLM systems—especially RAG and end-to-end machine learning pipelines. I’m comfortable across the full stack of model development, from fine-tuning and alignment to data engineering and production-style backends.

In my projects, I engineered a local multimodal assistant that integrates Speech-to-Text, Text-to-Speech, emotion analysis, and web-search-augmented retrieval (RAG). I also implemented an asynchronous FastAPI + WebSockets backend for low-latency token streaming and real-time emotional analysis, plus client-side intent routing to trigger search and voice responses dynamically.

I’ve worked deeply on scalable data workflows, including a RAG-based expert system for auditing Portuguese municipal financial reports by processing thousands of PDFs and building a high-precision ChromaDB vector database using semantic search. I also engineered a Python ETL pipeline extracting 438,000+ public-contract records from the BASE.gov portal, designing normalized MySQL staging and an OLAP star schema for multidimensional analytics.

My research and compute experience focus on LLM safety and high-throughput execution: I studied the impact of SFT and DPO on LLM safety and authored a technical report, and I’ve run parallel OCR workloads on the Deucalion supercomputer in Linux environments. I’m now seeking a junior AI/ML Engineer role in a fast-paced, international team where I can ship reliable AI systems.