I build production LLM and ML systems. Currently at Trinity Industries (Dallas, TX), where I ship Agentic AI-powered document intelligence and ensemble forecasting systems on Databricks, and pursuing my Ph.D. in Computer Science at The University of Texas at Arlington, where I research deep learning on graph-structured data.
🔗 ayonroy.com • LinkedIn • Google Scholar • 📫 royshouhag@gmail.com
LLM Document Extraction Pipeline — An end-to-end LLM + ML system automating multi-output repair-cost classification: Claude Sonnet clause extraction over 1,292 OCR-parsed lease contracts, paired with a CatBoost ensemble trained on 4M+ records, deployed to a Databricks/MLflow serving endpoint that scored 62,715 line items and replaced manual analyst review.
Ensemble Forecasting System — A 4-model ensemble forecasting railcar out-dates on Databricks: 60+ engineered features, KMeans segmentation with automated k-selection via knee detection, CatBoost classifiers, and per-cluster random forests producing confidence-bounded daily forecasts used by shop operations.
Cognitive Fatigue Detection (Research) — A Residual Graph Convolutional Network (ResGCN) detecting cognitive fatigue levels from gait-cycle data at UT Arlington.
Agentic AI & Developer Tooling (Side Projects) — Claude Code workflows (skills, sub-agents, hooks, MCP), agentic systems, and VS Code tooling. See claude-config.
LLM / GenAI
ML / Data
Languages
Full Stack
Cloud & Tools
