AI Solutions Architect — enterprise AI platform, governance, and enablement.
I turn manual, ad-hoc work into reusable systems other teams adopt, and I write the governance those systems run under. Twenty years in enterprise data management; the last two building production GenAI at a Federal Reserve trading and markets organization — three Microsoft Copilot agents in production, a Retrieval-Augmented Generation platform on AWS Bedrock, and an internal AI enablement platform now at 150+ users and 10,000+ sessions.
📍 Relocating to Fremont, CA · open to AI solutions architecture and enterprise AI roles
I direct coding agents against specs I own — Claude Code, Codex, Cursor, Kiro, Antigravity. The architecture decisions are mine; the keystrokes usually aren't. Everything below was built that way, and I can defend every design call in it.
| Project | What it is |
|---|---|
| security-audit | AI security scanner — 44 detection rules across secrets and OWASP Top 10, validated against GitLeaks, with LLM-based false-positive filtering. SARIF output for GitHub code scanning. Apache-2.0. |
| smart-chef-pantry-concierge | Multi-agent culinary concierge on Google ADK — Gemini 2.5 Flash, Vertex AI Agent Runtime, A2A protocol, A2UI components, RAG over a recipe corpus, deployed to Cloud Run. |
| MarketSounding | Multi-agent market intelligence platform. Enter a macro event — FOMC decision, tariff shock — and five primary-dealer AI agents debate it in real time. |
| fathom | Autonomous web research agent. Searches, extracts, and synthesizes sourced reports with full traceability and version history. |
| Launch-Neo4j-With-Terraform-and-Portainer | Neo4j Enterprise on AWS via Terraform and Docker, with Portainer and Caddy. A reproducible graph-database sandbox. |
| treasury-yield-summarizer | Treasury yield-curve dashboard with AI-generated market commentary. Next.js, React, shadcn/ui. |
- BeeQuill — local-first AI meeting notes. LangGraph retrieval agent with confidence scoring and query expansion, plus a custom int8-quantized vector store that cut embedding storage 4× (37 MB vs 146 MB per 100k segments) at sub-100ms retrieval. Eight LLM backends behind one adapter, an MCP server, 62 test suites.
- Table Tennis SB — iOS & Android match tracking for competitive junior players. Live on the App Store: four languages, in-app purchases, 750+ commits over 14 months. Built it because my son competes on the US National Juniors team.
- Whiskey Sensei — gamified prompt-engineering platform on AWS Bedrock with API Gateway, Cognito, and DynamoDB, provisioned through eight Terraform modules.
AI Agentic orchestration · RAG · AWS Bedrock · Google ADK · LangChain · LangGraph · MCP · vector databases Cloud AWS · Terraform · Databricks · Vertex AI · Cloud Run · Firestore Data Starburst · Neo4j · Collibra · semantic layer · metadata & lineage Languages Python · SQL · TypeScript · React