Perry Johnson

Hey, I’m Perry, an AI and data engineer. Most of my work has been in physical industries that run on messy data: heavy machinery, agriculture, supply chain and trade finance. I love working with excellent people to build products that solve meaningful problems.

I currently lead AI engineering at Digital Iron. Before that I co-founded Fishtail, and earlier I led data science at Farm Dog (acquired by Deveron).

Now

I live in Missoula, Montana. In my free time you can find me hiking, climbing, snowboarding and exploring wilderness areas. Lately I’ve been digging into regenerative agriculture, wildfire mitigation, supply chain efficiency and preventive healthcare.

Experience

Lead AI Engineer at Digital Iron

Building a scalable platform that ingests and enriches messy equipment service data, then powers agentic workflows and predictive analytics on top. It runs on resilient multi-model pipelines and a custom evaluation framework for improving the system.

ML Engineer and Co-founder at Fishtail

Built a document understanding engine to extract structured data from unstructured documents in any format
Developed a reconciliation engine to automate trade finance repayments
Created a global port health index to monitor risk across maritime supply chains

Lead Data Scientist at Farm Dog

Built data pipelines and in-field tools to support agronomists with real-time pest and disease insights.

Contract Projects

Built a configurable Python library to stress test hospital software by simulating patient flows across dynamic hospital layouts
Worked with a leading radiologist to build an image-segmentation model that maps colorectal anatomy and detects lymphoma lesions
Optimized SQL queries on an EV battery analytics platform to run 11x faster, unlocking real-time charging cost analysis

Quant Analyst at Strix Leviathan

Built and backtested algorithmic trading strategies across a wide range of cryptocurrency markets.

Personal Projects

Bird Scooter Nest Generator: A web app that predicts where scooters cluster into charging nests and recommends new drop-off spots, trained on a dataset I built from scooter locations across four cities
Reverse Engineering the Walk Score Algorithm: Built a dataset of ~7,800 Seattle locations and modeled Walk Score’s black-box walkability metric (R² 0.95) to find what actually makes a neighborhood walkable
Hiking Recommendation System: A personalized recommendation engine for hiking trails (“Netflix for hikes”) that learns from ~3,500 Washington hikes and 200k user reviews

Tech Stack

Languages: Python, Go, TypeScript, SQL
AI systems: LLM APIs (OpenAI, Anthropic, Groq, Fireworks), fine-tuning, RAG and agentic retrieval, eval frameworks
ML: PyTorch, scikit-learn, NumPy, Hugging Face, Weights & Biases
Data & pipelines: Airflow, Dagster, PostgreSQL, Redis, vector search (Weaviate, Meilisearch), PostGIS
Backend & infra: FastAPI, Django, Docker, Kubernetes, Terraform, GitHub Actions
Cloud: AWS, GCP, Supabase

Contact

If you’re working on something interesting or want to chat, reach out at perryrjohnson7@gmail.com.