AI engineer building LLM agent systems in production. Currently shipping voice AI agents for a US consumer lender, including a deployment handling ~2,000 automated calls/month across 49 branch locations.
- Voice AI agents on LiveKit Agents + Azure OpenAI (LLM/STT/TTS), running production call traffic end-to-end.
- A YAML DSL and runtime for conversation flows — nodes, guarded edges, intents, terminal outcomes, validated at startup — so a new call scenario ships as an isolated package.
- Dispatch services: FastAPI, Pydantic v2, Azure Service Bus (immediate and scheduled), event-driven drain jobs, JWT auth, PCI/PII rejection at the API boundary.
- MCP servers exposing enterprise data to AI assistants, and evaluation harnesses that grade agent behaviour — golden-case corpora, tool-call intent graders, refusal and faithfulness graders, promptfoo, LLM-as-judge conversation tests.
AI & agents — LiveKit Agents, Azure OpenAI (LLM/STT/TTS), MCP, promptfoo, LLM-as-judge evaluation Backend — Python, FastAPI, Pydantic v2, Azure Service Bus Cloud & delivery — Azure Container Apps, Bicep, Azure DevOps pipelines, OpenTelemetry / Application Insights Testing — golden-case evaluation corpora, tool-call intent graders, refusal & faithfulness graders
- Claude Certified Architect – Foundations — Anthropic Education
- Claude Partner Badge — Claude Code
- AWS Certified Solutions Architect – Associate — AWS
- AWS Certified Developer – Associate — AWS