Building intelligent
data systems &
published research

Compares three LLM agent deployment patterns — Managed Agents, DIY Factory (Python + GCP), and Direct API — across 45 controlled measurements on financial data governance tasks. Reveals that prompt caching inverts the expected cost hierarchy: DIY accumulates 21.68x context tokens while managed caching reduces costs by 90%. Includes security analysis for regulated fintech (PCI-DSS, SBS Peru, Ley 29733) with a practical decision framework for data sovereignty.

Introduces the GRAFID framework with three novel metrics (Feature Richness Index, Graph Signal Gain, Cost-Effectiveness Ratio) to determine when GNNs outperform XGBoost for fraud detection. Evaluated on IEEE-CIS (590K transactions) and Credit Card EU (284K) datasets with 20+ model configurations, statistical validation (Wilcoxon, McNemar, Cohen's d), and multi-seed reproducibility.

Demonstrates that data quality improvements (ground truth correction, embedding deduplication) yielded a 27% improvement in Precision@5 with zero code changes and zero cost, outperforming neural reranking approaches. Based on 28 formal evaluations with 52 reference questions.

Design and prototype of an autonomous two-wheeled mobile robot using inverted pendulum control systems, with stress simulations and material validation in Autodesk Inventor.