Build production AI systems spanning real-time training, agent lifecycle management, evaluation, and enterprise retrieval.

• Shipped a real-time AI call-center training platform with configurable voice roleplays, reducing new-hire ramp-up time by 28%.
• Designed an LLM-as-judge system calibrated on human-scored transcripts for reliable coaching feedback.
• Built an AI-agent lifecycle control plane with versioned registry metadata, policy gates, durable evaluation runs, GitLab promotion, and production observability.
• Architected a planner-executor-judge evaluation harness with holdout suites for quality, safety, latency, and cost regressions.
• Engineered a Vertex AI Search workspace with metadata-filtered retrieval, configurable chunking, and citation-linked answers for private datastores.

Built retrieval, model access, and multimodal capabilities for internal enterprise AI products.

• Developed a Flask RAG system with embeddings and Gemini for source-grounded policy answers and document summaries, cutting search time by 70%.
• Engineered a fault-tolerant internal LLM API gateway with automatic retries and structured request logging.
• Built a multimodal React and FastAPI assistant on Vertex AI for text, PDF, image, and voice workflows on a platform handling 6M+ annual prompts.