At BMC Software, I built anomaly-detection workflows for BMC Helix Operations Management and event-correlation pipelines that helped surface likely causes, saving roughly 15–30 minutes per incident investigation compared with manual triage on sample incidents.
At Stony Brook University, I fine-tuned Mistral-7B with QLoRA and PEFT, reducing hallucinations by 34%. I also built an evaluation framework comparing fine-tuned models, base LLMs, and RAG pipelines.
In personal projects, I built a financial intelligence RAG platform and a multi-agent assistant combining retrieval, predictive risk scoring, and web research. My work spans Python, machine learning, and LLM applications.
