Architected an agentic AI system that autonomously diagnoses RAN network alarms end-to-end, cutting field-engineer diagnosis time from ~2 hours to 10-20 minutes. Designed a plan-and-execute agent (LLM planner, ReAct execution loop, verification) with confidence gating and human-in-the-loop escalation. Built a RAG pipeline over a LanceDB knowledge base of alarm runbooks with local LLM inference on Ollama across on-prem NVIDIA A30 GPUs, and a read-only MCP server exposing 100+ network diagnostic tools to LLM agents. Also designed an in-platform AI chatbot for the Optimus (Django + React) system used by field engineers.

[001]

Analyzed theater demographic data across 300+ locations to identify trends in food and beverage performance. Built K-Means clustering models to segment theaters and compared KPIs like per capita spending and occupancy across clusters. Did market basket analysis to find high-value product pairings and improve combo recommendations. Also worked on the 'Pop & Top' initiative using guest surveys and international sales data. Designed executive presentations in Power BI and matplotlib to translate complex findings into actionable insights.

[002]

Developed REST APIs in C# ASP.NET for tax form validation that cut processing time by 10-15%. Refactored validation logic into modular components, reducing turnaround from weeks to near real-time. Queried large datasets in SQL Server and converted legacy XML rules into modern XSLT for better integration. Collaborated with a global Agile team across Canada, India, Mexico, and Russia.

[003]