At SIDIA Institute of Science and Technology, I build and maintain interconnected internal services used by more than 300 employees. I cut API response times from several seconds to under one second by refactoring FastAPI database connection handling, and added caching with background refresh and MinIO storage.
I built a FastAPI and SQLAlchemy reporting service that schedules test dates for around 70 active product projects, sends daily reports, and serves several internal systems. I also built semantic search over issues using pgvector and Faiss, with RabbitMQ handling embedding updates.
At SIDIA, I built an internal LLM chatbot using a self-hosted gpt-oss-120b, tool calling, and specialist subagents to analyze container status and logs. I also migrated legacy Flask services to FastAPI back ends with React front ends and took part in moving systems to centralized authentication.
For my B.Eng. thesis, I built a semantic job recommendation API with Sentence-BERT embeddings stored in pgvector. Its hybrid model improved nDCG@5 from 0.075 with TF-IDF to 0.133; I also built Liga pra Mim, an AI phone helpline, as a solo hackathon project.
