At Proviloops, I architected a multimodal LLM content-intelligence platform using text, image, and graph data. It improved classification precision by 37% and reduced moderation latency by 42%.
I also designed fraud and policy-risk detection systems for financial and insurance use cases, cutting false alerts by 33%. For insurance fraud claims, I developed a graph-based entity-resolution pipeline that reduced manual review hours by 70% and saved $1.8M annually.
At CogniWorks AI, I designed real-time inference architecture processing 500M+ daily events, with sub-30ms latency at peak load. I also modernized MLOps workflows, reducing deployment cycles from two weeks to under four hours across 12 production models.
Earlier, at Techionik, I developed machine-learning models for churn prediction and lifetime-value forecasting, achieving 82% recall across 4M+ customer records. I also built WeFly, an aviation and fleet-management platform that reduced manual work by 70% and operating costs by 40%.
