Building Production AI Systems

At Hyland I build agentic AI that powers enterprise document workflows for Fortune 500 companies. Users describe what they need in plain language — the agentic layer interprets intent, reasons over available tools via function calling, and generates low-code configurations without manual wiring. Document intelligence pipelines handle categorization, extraction, and classification at scale, with confidence scoring and human-in-the-loop checkpoints where the stakes are high.

Outside of Hyland, I built a production LLM proxy server that unifies seven providers — OpenAI, Anthropic, Azure AI Foundry, AWS Bedrock, Groq, and local models — behind a single .NET Core API. It tracks spend in real time and routes each request to the most cost-effective model for the task. The same cost-optimization thinking applies to every RAG system I design: right chunking strategy, right retrieval depth, right model tier.

My RAG work ranges from the ChromaDB-backed vector search powering this portfolio chatbot to pgvector integrations on Postgres for production systems. I approach vector databases as infrastructure — choosing between them based on operational requirements, not hype. The result is AI features that ship, scale, and stay within budget.