Shinkiro
Turn an idea into a working web app. Describe what you want to build, shape a first version, and try it with real people. I’m opening Shinkiro to early design partners.
Landon GraySelected work
Production AI, evaluation, and helping people use what we build.
My software consulting background connects the technical work to the people and processes around it. These examples show that approach in practice.
Making an investor-verification agent observable and testable
DealMaker · 2025–26
I migrated an investor-verification agent from n8n to Google ADK, comparing behavior in shadow mode before cutover.
My contribution: I built the evaluation workflow with domain experts reviewing traces and annotating failures, and made runs searchable and replayable in Datadog. End-to-end latency stayed under ten seconds.
The approach: Compare behavior before switching systems, and give the people who understand the work a practical way to identify failures.
DealMaker Labs: an internal AI tools program
DealMaker · 2026
My work on the operating model connects employee ideas, build support, production hosting, governance guidelines, and cost tracking.
Early adoption: Seven employee-built tools were registered across Sales, Marketing, Operations, and Broker Dealer in the first week.
The approach: Give people a path from an idea to a tool they can use, with support and ownership beyond the initial build.
AI search for vehicle and RV dealerships
Automotive · 2024–25
I built a retrieval-augmented search system that lets shoppers query inventory conversationally.
My contribution: I replaced subjective quality checks with automated RAGAS evaluation as a release gate and profiled the pipeline to reduce response times.
The approach: Make quality a repeatable check during delivery, and measure where the system spends its time.
Helping engineers build AI-powered search
Healthcare · 2024
I trained 14 healthcare engineers in NLP, vector search, language models, and retrieval-augmented generation, with regulatory considerations included in the material. I then open-sourced the course.
Course materials · Training labs
What I’m building
FoodieLog
A home for the dishes you want to remember, from restaurant discoveries to your own kitchen. Save photos and notes, collect your favorites, and share food guides with friends.
Open-source work
- Google ADK for Ruby — an unofficial Ruby port of Google’s Agent Development Kit.
- Forecasting the Future — a weather-prediction model in Ruby, presented at RailsConf 2023.
- SimpleRAG — a small Ruby implementation that makes the stages of retrieval readable.
- Ruby ReAct Agent — a direct implementation of the reason, act, observe loop.
Have a related challenge? See my advisory focus, or email me about an opportunity.