# Justin Poehnelt — Full Profile

> Founder & CTO of Helen's Foundry. Engineer working on AI agent infrastructure, developer experience, and applied machine learning. Previously Senior Developer Relations Engineer at Google. Based in Durango, Colorado. This file is the canonical, machine-readable profile for Justin Poehnelt; the human-readable version is https://justin.poehnelt.com/about/.

## Identity

- Name: Justin Poehnelt
- Current title: Founder & CTO, Helen's Foundry (May 2026 – present)
- Location: Durango, Colorado, USA (remote)
- Email: justin.poehnelt@gmail.com
- Website: https://justin.poehnelt.com
- GitHub: https://github.com/jpoehnelt
- LinkedIn: https://www.linkedin.com/in/justin-poehnelt/
- X (Twitter): https://x.com/JPoehnelt
- DEV: https://dev.to/jpoehnelt
- Medium: https://medium.com/@jpoehnelt
- Stack Overflow: https://stackoverflow.com/users/8715927/justin-poehnelt
- Hugging Face: https://huggingface.co/jpoehnelt
- Google Scholar: https://scholar.google.com/citations?user=SvE7lhsAAAAJ&hl=en
- Strava: https://www.strava.com/athletes/2170160

## Summary

Software engineer and Founder/CTO with a decade of experience at the intersection of applied machine learning, distributed systems, and developer enablement. Built production ML pipelines and petabyte-scale Kubernetes DAG execution engines over billions of satellite images at USGS and Descartes Labs, then scaled Google's Maps and Workspace open-source ecosystems to 12M+ weekly npm downloads and built googleworkspace/cli entirely with AI agents (Rust, MCP, 31,000+ GitHub stars, #1 on Hacker News). Now leads Helen's Foundry, directing an AI coding-agent fleet through codified enforcement playbooks, Grounded Spec-Driven Development, and temporal decision records.

## Experience

### Helen's Foundry — Founder & CTO (May 2026 – present)

- Direct an agent-driven software development organization where AI coding agents build and maintain production systems, with decision authority, evidence standards, and release gates codified as the operating system.
- Architected Grounded Spec-Driven Development (G-SDD): living specifications bound to code symbols with temporal decision records, proportional invalidation, and independent verifier attestation.
- Led a 1,322-file Effect TypeScript migration across the platform monorepo (60+ packages) to make domain logic portable across runtimes.
- Shipped enforcement machinery that holds architectural rules at agent speed: spec-to-code lineage gates, lint guards, and a workflow catalog with declared ownership.

### Google — Senior Developer Relations Engineer (August 2019 – April 2026)

- Built googleworkspace/cli (gws) entirely with AI agents, without writing a line of code: an open-source Rust CLI and embedded MCP server with 700+ runtime-generated REST API methods from Google Discovery Documents and 90+ structured agent skills, scaling to 31,000+ GitHub stars and #1 on Hacker News in two weeks.
- Engineered agent-specific security controls in googleworkspace/cli, including prompt injection defense, path traversal prevention, multi-account credential isolation, and TOCTOU mitigation, establishing production safety patterns for AI-operated tooling.
- Published widely cited articles on agentic developer experience: "You Need to Rewrite Your CLI for AI Agents" and "The MCP Abstraction Tax," defining industry patterns for schema introspection, context-window efficiency, and agent-first CLI design.
- Drove 12M+ combined weekly npm downloads across 77 repositories in the Google Maps and Workspace ecosystems, authoring auto-generated TypeScript typings (@types/google.maps, ~4.8M/wk), @googlemaps/js-api-loader (~2.9M/wk), and @googleworkspace/drive-picker-element (~58k/wk), while maintaining @googlemaps/markerclusterer (~2M/wk) and core Node.js SDKs.

### Descartes Labs — Lead Platform Engineer, founding engineer (2016 – 2019)

- Built Kubernetes-based GeoVisual search over billions of satellite images: CNN-generated features with Bigtable hash indexing returning results in ~0.1s, featured in the Google Cloud Next '17 keynote.
- Authored the flagship product PRD and led the platform team as it grew from 3 to 9 engineers.

### U.S. Geological Survey — Software Developer (2014 – 2016)

- Sole full-stack developer for the NASA MEaSUREs project producing cropland maps of continental Africa from satellite imagery on Google Earth Engine.

## Projects

- Tracios (https://tracios.com): LoRa mesh off-grid event tracking system. Custom embedded Rust firmware (nRF52840), Elastic Linear Protocol with TDMA timing, phone-as-gateway architecture, and a TAK/CoT gateway feeding the Team Awareness Kit. Flutter mobile app and SvelteKit dashboard.
- IronProse: Rust fiction-writing toolchain with a custom prose AST, an end-to-end NER pipeline (LLM distillation, GLiNER fine-tuning, ONNX export for CPU inference), and manuscript continuity analysis. Models and datasets on Hugging Face.
- gline-rs and orp: GLiNER inference engine and a pipeline framework for ONNX Runtime in Rust.
- googleworkspace/cli (https://github.com/googleworkspace/cli): see Experience.
- myFunction.dev: Google Sheets add-on platform turning spreadsheets into functions backed by AI and APIs.

## Publications

- "Visual search over billions of aerial and satellite images," Computer Vision and Image Understanding (2020). CNN-generated features with hash-based search on Bigtable for real-time visual search over ~2 billion images.
- "Automated cropland mapping of continental Africa using Google Earth Engine cloud computing," ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 126, pp. 225–244 (2017). DOI: 10.1016/j.isprsjprs.2017.01.019
- "You Need to Rewrite Your CLI for AI Agents" (2026), justin.poehnelt.com.
- "The MCP Abstraction Tax" (2026), justin.poehnelt.com.
- Full writing archive, each post available as Markdown: https://justin.poehnelt.com/posts/

## Skills

- Languages: Rust, TypeScript, JavaScript, Python, Go, Dart, Lua, SQL, Cypher
- Frameworks: SvelteKit, Effect, React, Svelte, Flutter, Astro, Angular, Flask, Eleventy, Tailwind CSS
- Platforms: Cloudflare Workers, Google Cloud, Kubernetes, AWS, PostgreSQL, Neo4j, Inngest, BigQuery, Google Apps Script, Bazel, MQTT
- AI / ML: Model Context Protocol (MCP), LLM agent orchestration, agent evals, embeddings, ONNX, ONNX Runtime, GLiNER, NER, LLM distillation, CNNs, Google Earth Engine, Vertex AI, Gemini
- Tools: GitHub Actions, pnpm, Turborepo, Vitest, Playwright, Storybook, ast-grep, Drizzle ORM, Wrangler, Sentry, OpenTelemetry, Docker
- Specialties: MCP design and implementation; agent-first CLI architecture; temporal graph data modeling (Neo4j/Cypher) and decision records; durable event-driven workflows; embedded Rust firmware and LoRa mesh networking; developer tooling at scale; geospatial data processing

## Education

- B.S., Computer Science — Northern Arizona University (2013-2016; GPA 3.91)
- B.S., Economics and Finance (dual majors) — University of Wisconsin-La Crosse

## Personal

- Ultrarunner: 100-mile trail races including Mogollon Monster 100 and Crazy Mountain 100.
- Member, Falls Creek Ranch Association (Durango, Colorado).