At Freelance, I ship LLM systems for international clients and my own agency, including an outbound email platform, a multilingual article pipeline, and an omnichannel sales agent. I build evaluation harnesses to measure quality before claiming results.
I built AI Persona, a voice assistant with a real-time avatar that can run its owner’s Gmail, Calendar, Contacts, and Drive by voice. In a 552-request evaluation, improvements raised its pass rate from 90% to 98%.
For Vendra, I replaced single-shot retrieval with agentic RAG, raising answer accuracy from 58% to 92% and reducing hallucinations from 8.6% to 1.2% on an 81-question evaluation. I also shipped an OAuth 2.1 MCP server that lets outside agents access the sales agent securely.
