As an independent Machine Learning Researcher & AI Engineer, I implemented a training-free way to steer LLaMA 2 outputs through activation engineering. I also built an evaluation harness to compare steered and baseline responses, check robustness, and log generated answers.
I built an autonomous AI sales assistant with job ingestion, qualification, proposal generation, and outcome tracking. Its worker-based architecture combines a Streamlit interface, browser automation, and OpenRouter LLM integration.
For LizaAlert Detector, I delivered a YOLOv11 pipeline for high-resolution aerial imagery and served inference through a Flask REST API integrated with a Telegram bot. Earlier, I designed a corporate DWH handling about one trillion records across more than ten years of history, helping cut mortgage-loan securitization time from six months to two.

