Labs
Client work stays private. Ideas I want to think through end up here, in the open, running ongithub.com/mrskinzo . Use them, fork them, tell me what you'd do differently. Prototypes are labeled prototypes.
Three retrieval configurations run against the same help-center corpus with one eval harness: chunk size and overlap swept across dense retrieval, comparing cosine similarity against MMR. The numbers picked the winner, not the demo.
Companion to the RAG case study
A reproducible data study behind the essay "Ecommerce AI Hit an Inflection Point." It pulls thousands of real GitHub projects, runs topic modeling and a text classifier over them, and charts the trajectory. Plain Python, no API key.
Quiet category to a crowd: ~20x in two years
A LangGraph prototype where a coordinator agent routes work to research, analysis, and writing agents, with hybrid RAG and an output verification pass. Built to think through multi-agent routing, not to run your production workload.
Prototype · architecture study
A small Python tool that runs the same question through different prompt strategies (direct, structured, chain of thought) and lays the results side by side with timing and token counts. A quick check before a real eval harness exists.
Small utility · Python
Work with me
Hiring a firmA few consulting engagements each quarter throughIntelligent CX Consulting . Start at services.
Hiring a personOpen to conversational AI, AI product, and applied AI roles. The resume is at /resume.
How can I help you, today?