Applied-AI engineer. I build Model Context Protocol platforms, agentic workflows, and the evaluation harnesses that keep them honest — then ship them as public, running systems.
Data Analyst (Analytics Engineering) at Teleparty by day; independent AI-systems builder the rest of the time.
| Project | What it is |
|---|---|
| studio-mcp · live demo → | 14-tool MCP server running a generative pipeline end to end (stdio + streamable-HTTP). Ships with a cinematic web console, an LLM Director that drives the tools autonomously, a multi-agent orchestration graph, and Langfuse-traced eval — QC gate + human-in-the-loop throughout |
| creative-rag | Hybrid RAG (dense + BM25 + rerank) with a citation-verification eval harness wired into CI — the build fails on unsupported claims; Docker + Terraform for AWS ECS |
| sf-flow-demo | Intelligent support automation on Salesforce Service Cloud — record-triggered triage flows + an LLM agent that drives them over the REST Actions API, verified live against a real org |
| insider-signals | 7 scheduled AI agents read public US-gov + on-chain data, vote on consensus, and alert when ≥3 agree — GitHub Actions + Supabase |
| fusion | Mixture-of-agents LLM gateway; benchmarked to match frontier quality at ~28% lower cost — with the honest failure case published too |
| knowledge-os | Obsidian LLM-wiki run by Claude Code + graphify knowledge-graph tool + a Telegram agent over a capability registry |
| snowflake-port · finance-os | Athena→Snowflake analytics patterns (ran live on a trial) · agentic portfolio OS with an LLM advisor that argues from my own written rules (public mirror; data private) |
MCP · agentic workflows · LLM eval / LLM-as-judge · RAG · Salesforce Flow · Python · TypeScript / Next.js · Snowflake · Docker · GitHub Actions · Terraform · AWS · Cloudflare Workers · PyTorch