Production GenAI systems I've designed and shipped end-to-end — from agentic platforms and RAG pipelines to voice AI and the LLMOps that delivers them. Most are proprietary; case studies describe the architecture and decisions without internal detail.
A document backend where SQL-queue workers claim jobs with SKIP LOCKED, heartbeat while they OCR, chunk, and embed into pgvector — and recover cleanly when a worker dies mid-job.
A chat interface over spreadsheets — each question is classified by intent, answered with pandas code generated and run in a sandbox, and returned as a chart or table instead of a hallucinated number.
An ops dashboard for the conversational AI products — browse real conversations, run LLM-as-judge evaluations, and manage per-firm bot configuration with diff-review and an audit log.
A marketing-site chatbot that qualifies visitors in conversation, writes leads into the CRM, and books product-demo slots straight into the sales calendar.
A drafting service that OCRs case files in batch, builds structured medical overviews with a team of extraction agents, and drafts legal documents from reusable blueprint templates.