Joshua Park

Architected an enterprise AI document intelligence platform with Python/FastAPI that automated regulated, multi-step agentic workflows with full audit traceability. Built LangGraph multi-agent tool-calling systems and scalable Azure OpenAI RAG pipelines, cutting compliance manual review time by ~40% while deploying HIPAA-compliant workflows and secure Kubernetes CI/CD.

Prototyped Claude-based autonomous agentic workflows using LangGraph for multi-turn document retrieval and reasoning. Optimized production RAG serving with Ray Serve to reduce p99 latency by 35%, implemented guarded PII-filtered architectures, and delivered red-teaming that improved model safety scores by 28%.

Developed distributed training pipelines on Databricks (AWS) processing 100M+ fraud records, enabling real-time risk scoring with sub-100ms inference latency. Built SageMaker Pipelines with MLflow-governed model lifecycles and observability, and implemented graph-based feature generation with Neo4j for anomaly detection.

Built clinical text classification and NER models using scikit-learn and spaCy, improving downstream analytics accuracy by 18%. Implemented OCR pipelines with OpenCV/Tesseract achieving ~95% text extraction accuracy and developed FastAPI endpoints and PII-filtering for compliant model training and inference.

Developed fraud detection models (e.g., gradient boosting, random forest, logistic regression, XGBoost) and improved fraud precision by 22% through feature engineering and hyperparameter optimization. Built large-scale ETL pipelines with Hadoop/Hive/Spark/SQL and graph-based account linkage in Neo4j, reducing false positives by 15% via tuning of scoring thresholds and review workflows.

Earned an MBA with a focus in Finance.

Earned a B.S. in Mathematics and Computer Science.