At REZINX, I fine-tune language models with LoRA for production use cases and evaluate embedding models for job classification. I also built a recommendation engine using Superlinked and a Grafana monitoring dashboard.
At REZINX, I developed a computer vision hazard detection system to identify fires, knives, and stairs and assess their proximity to children.
For my QuestionPrep project, I built a full-stack AI interview platform using FastAPI, llama.cpp, and ChromaDB. I also implemented in-browser stress detection with a TCN model exported to ONNX Runtime Web and a custom JavaScript DSP pipeline.
My projects include a profanity detection and rewriting system combining LoRA-fine-tuned DistilBERT with a QLoRA-fine-tuned LLM, and a network intrusion classifier that reached 94.67% accuracy using ensemble techniques.
