At Accordion, I built demand forecasts for 12,000+ SKUs across PE clients, cutting MAPE by 15%. Feeding the forecasts into demand and procurement planning helped cut inventory holding costs by $4.2M annually.
I also built anomaly-detection models covering $350M+ in inventory across 28 warehouses. Python and SQL ERP pipelines I built process 50M+ monthly records, while automated ETL and quality control cut manual overhead by 40%.
At DeepVision, I built a multi-agent assistant that helped retail managers query store performance, inventory, and root causes across 45+ stores. It combined LangGraph agents, Neo4j knowledge graphs, and multimodal data to answer 100+ queries at 94% accuracy and cut root-cause analysis time by 60%.
On my Domain-Specific Customer Support LLM project, I fine-tuned Llama 2 7B and built a RAG pipeline that improved answer relevance to 94%. I also worked on a property-rentals project analyzing guest-host chats, identifying 12+ service issue clusters and automating responses to cut query resolution time by 35%.
