Building an LLM-based QA-automation system for customer-service operations, plus an ARIMA forecasting model turning historical data into strategic projections with confidence intervals. Researched LLM observability and evaluation with Langfuse, including an agent error taxonomy and an LLM-as-judge prompt for trajectory-level failure detection.
One-on-one Python tutoring for university Economics and Physics students, reinforcing programming fundamentals and problem-solving techniques.
Built a local-LLM knowledge and compliance assistant using Retrieval-Augmented Generation to support organisation-wide quality management, plus AI-driven document validation and process-compliance workflows.
Built custom OpenCV pipelines for interactive visual systems: color filtering, homography estimation, foreground separation, camera calibration and pose estimation.