At ABN AMRO, I built an LLM-powered virtual assistant used by 22,000+ employees. It combined intent recognition, confidence-based routing and embedding-based RAG, and I evaluated it with A/B tests and NPS tracking.
I also built the Azure side of the platform during the AWS-to-Azure migration, including its integrations and testing framework. The cutover happened with zero production incidents, and two AI components went into production for other teams to reuse.
As an Independent ML Engineer, I built a Claude-powered multi-agent content-generation pipeline, orchestrated in n8n and connected to Telegram and PostgreSQL. I also open-sourced a Python MCP server that helps evaluate creative choices against view counts.
Earlier, at Netflix, I built a conversational incident assistant that cut the response SLA from 10 minutes to 1 minute. Across my work, I take ML systems from data and model evaluation through to production, using short research-then-build cycles.

