AI Engineer · Builder · Ships to Production
RAG assistants, data pipelines, full-stack apps. I build the AI systems that teams actually use every day.
Selected Work
Dragged a company's AI capability forward — from zero to daily production usage across three teams.
Custom RAG assistant in production. 69.2% accuracy, ~30 msgs/day, ~30 min saved per user. Evolving from custom RAG to MCP.
Bronze/Silver/Gold data architecture turning messy HVAC supplier catalogs into clean, import-ready costbooks. Deployed on EKS.
PRD through delivery: 15-module React/TypeScript wizard with Supabase backend and Salesforce integration. Owned the full product cycle.
Internal system for managing and versioning prompts across teams.
LLM-powered grading pipeline for sales certification assessments.
Multi-agent pipeline: Gong transcripts to tailored onboarding curriculum.
Claude Code plugin for semantic memory and document retrieval.
What I Do
RAG assistants, eval frameworks, agent pipelines — built, deployed, and used daily by real teams.
Bronze/Silver/Gold architecture for turning chaotic supplier catalogs into clean, structured data.
From PRD to shipped product. 15-module React/TS wizard with Supabase + Salesforce integration.
Claude Code, agent teams, MCP plugins — using the latest AI tooling to ship faster, not just talk about it.
I don't build demos. Juju runs in production at FieldPulse with 69.2% accuracy and ~30 daily messages. I own the full loop: build, deploy, instrument, iterate.
Proof: Juju case study →
HVAC manufacturers ship catalogs as PDFs, spreadsheets, and HTML. The Costbook ETL pipeline normalizes all of it into a clean Bronze/Silver/Gold architecture, validates against AHRI data, and deploys on EKS.
Proof: Costbook ETL case study →
I wrote the PRD, designed the architecture, built all 15 modules, integrated Supabase and Salesforce, and shipped the Customer Onboarding App. No handoffs, no waiting on other teams.
Proof: Onboarding App case study →
Prompt Library, Sales Cert grading, Gong automation, GoodMem plugin — I build with Claude Code and AI-native tooling daily. Each project ships in days, not months.
Proof: Prompt Library, Sales Cert, GoodMem →
Juju runs in production with 69.2% accuracy and ~30 daily messages. I own the full loop: build, deploy, instrument, iterate.
Bronze/Silver/Gold architecture turning chaotic HVAC supplier catalogs into clean, structured data on EKS.
PRD to shipped product. 15-module React/TS wizard with Supabase + Salesforce. No handoffs.
Claude Code, MCP plugins, agent teams — each project ships in days, not months.
69.2%
Juju Accuracy
~30
Messages / Day
~30 min
Saved / User / Day
3
Teams Using Juju Daily
Building an AI system? Need someone who ships, not pitches? Let's talk.