At Spindox Labs, I own a production probabilistic forecasting service for smart-meter fleets, using per-device and per-community LightGBM models with conformal prediction intervals. I also maintain the Prefect and dbt pipelines that feed the service.
I built an open-source forecasting engine and reproducible benchmark comparing gradient boosting with time-series foundation models. For a grid operator, I built an explainable weather-driven fault early-warning model from outage records and LLM-classified fault reports.
Previously at Be Consulting, I built a retrieval-based question-answering system over regulatory documents and a computer-vision pipeline for detecting document alterations. At Deloitte Consulting, I developed transaction anomaly detection that cut investigation time from three hours to one per cycle.
