At EAGE Technologies, I built and benchmarked LightGBM and LSTM forecasting models for solar power output, scoring 89% and 80% on evaluation. I also designed retraining triggers and performance-ratio-based drift monitoring.
I architected an agentic decision pipeline that combined forecasts, rule-based risk logic, and LLM function calling to generate operational directives and risk alerts. I also built a real-time computer vision pipeline using YOLOv8, InsightFace, and FAISS, connected to STM32 hardware for a client defense application.
For two pilot locations, I designed an IoT-to-cloud soil intelligence engine that used cold-start threshold logic and location-based climate APIs without historical data. I worked with client stakeholders and engineering leads to translate operational needs into specifications and deploy pilot systems.
Recently, I built EV Ops Copilot, an agentic RAG assistant for EV operations and sales questions. It combines hybrid search and grounded cited answers with SQL guardrails, monitoring, cost controls, and regression evaluation.
