Engineering applied LLM systems and robust backends.
Currently shipping coaching AI for sales agents at NovumAI.
About
I build production AI systems and the backend infrastructure that runs them. Currently the primary AI/ML engineer at NovumAI, shipping real-time coaching AI for sales agents — from raw data pipelines and model training through to live deployment.
Recent highlights: a custom AI model delivering 42% better coaching quality and 41% lower response latency, an intelligent onboarding platform that cut ramp time from 5 days to 2, and a Claude Code marketplace plugin used by AI-assisted developers worldwide. I build things that ship, scale, and hold up under real load.
What I Work On
- LLM Fine-Tuning & Alignment
- RAG Pipelines
- ReAct Agents
- LLM Evaluation (LLM-as-Judge)
- ML Data Pipelines
- Backend & APIs
- AWS & Cloud Infrastructure
- Vector Search (Pinecone)
- Distributed Systems
- Python & TypeScript
- PostgreSQL · DynamoDB · Redis
- Docker & CI/CD
Selected Work
View All Projects →
Software Engineer (primary AI/ML engineer) · Nov 2025 – Present
NovumAI — Coaching AI for Sales Teams
Building real-time coaching AI for sales agents on live calls. Trained a custom model that improved coaching quality by 42% and cut response latency by 41%. Architected the data pipeline, evaluation harness, and production deployment.
- AWS Bedrock
- SageMaker
- Llama 3.1
- Python
- DynamoDB
- Redis
Visit Site →
Claude Professor
Claude Code marketplace plugin
A Claude Code plugin that embeds spaced-repetition concept teaching into AI-assisted development to combat "vibe coding." 407 concepts across 18 domains; ~88 automated tests; zero external dependencies.
- Node.js
- FSRS-5
- Claude API
View on GitHub →
Sherpa
Multi-tenant Slack volunteer onboarding bot
Originally built for Changing The Present, then generalized into a distributable platform any nonprofit can install via Slack OAuth. ReAct agent with 5 tool types, 4-layer cost protection, $0.50–$2.50/month per workspace.
- Python
- Gemini
- Pinecone
- AWS Lambda
- DynamoDB
View on GitHub →