Michael Demilio

Engineered full-stack AI apps combining React/Angular frontend with Python FastAPI backends; improved data quality and reliability by designing ingestion and transformation pipelines for knowledge-grounded LLM features; provided modular dev/staging/prod environments using Terraform for reproducible deployments; introduced structured logging, health checks, and Prometheus/Grafana monitoring for production visibility; collaborated with cross-functional teams to ensure performance, security, and compliance.

End-to-end LLM observability: prompt regression testing, hallucination monitoring, and cost telemetry; reduced costs by ~42% through prompt optimization, Redis caching, and cross-endpoint routing across Bedrock and OpenAI. Document extraction pipeline and data validation/processing frameworks; standardized containerized deployments (Docker, Kubernetes/EKS, Terraform) for reproducible environments; collaborated with frontend, product, and data teams to translate AI capabilities into business value.

Enterprise GenAI knowledge platform: built a production-grade multi-agent RAG platform for 200+ daily users; persistent memory with LangGraph orchestration; integrated Bedrock Agents with custom action groups for live data retrieval, downstream API execution, and human-in-the-loop validation. End-to-end LLM observability and cost governance framework using LangSmith, Arize Phoenix, and CloudWatch; cut inference costs ~42% via prompt optimization, semantic caching (Redis), and intelligent model routing across Bedrock and OpenAI endpoints. Document extraction pipeline (DOCX, PDFs, scanned images) using Azure Cognitive Services, Hugging Face transformers, PyMuPDF; deployed services to AWS via Helm CI/CD pipelines; Azure Functions handled continuous knowledge-base updates. Structured deployments (Docker, Kubernetes/EKS, Terraform) to accelerate releases and improve rollback reliability. Collaborated with frontend, product, and data teams to translate AI capabilities into measurable busines