Work
Case studies for work under NDA, open-source AI tooling, products I built or helped build, and writing.
Browse the full projects library, including the BYU AI Foundry's work and paid client work.
Showing all 30 items
-
Case study · Summer 2026
Siemens: an AI agent platform for the US workforce
Built and launched Spark AI, a 120+ agent platform for 1,000+ Siemens US employees across every major function, took adoption from 3% to 65% in 10 weeks, and presented a 3/6/12-month AI roadmap to US leadership.
Role: AI Product Strategist, XPS Program
at Siemens
Read case study: Siemens: an AI agent platform for the US workforce
-
Case study · 2022 to 2025
National Grid: a cloud and AI platform for a siloed enterprise
Owned the cloud and AI/ML roadmap at a 20,000-person utility: launched a RAG-based generative AI platform to 3,000+ monthly users and grew an internal cloud services platform from $2M to $20M+ in annual chargeback.
Role: Principal Product Manager
at National Grid
Read case study: National Grid: a cloud and AI platform for a siloed enterprise
-
Case study · 2020 to 2022
Deloitte: machine learning for Fortune 100 clients
ML and data work on Fortune 100 engagements: an ML contact-center product for Pfizer that cut average queue time from 75 to 15 minutes, and a clustering model for GM that saved $17.5M.
Role: Machine Learning Consultant
at Deloitte
Read case study: Deloitte: machine learning for Fortune 100 clients
-
Case study · 2018 to 2020
Brio Energy: sales and accounting systems for a solar company
Built a custom CRM that cut the sales-to-accounting turnaround from 5 days to 15 minutes, a marketing site producing 175+ qualified leads a week, and microservices that cut accounting spend by $745K a year.
Role: Senior Full-Stack Developer / Product Owner
at Brio Energy
Read case study: Brio Energy: sales and accounting systems for a solar company
-
Writing
Nerve Center teardown
A public teardown of the multi-agent system I run: the architecture, three production failures with root causes, and the fixes.
Role: I built and run the private system and published the teardown and the architecture map.
-
Open source
mcp-judge
A calibrated multi-persona LLM judge exposed as an MCP server, shipped with an eval harness that proves the calibration.
Role: I built it, open-sourced through the BYU AI Foundry.
-
Open source
claude-deploy-kit
Enterprise controls for deploying Claude agents: a policy chokepoint, destructive-command deny hook, eval gate and redacting audit log.
Role: I generalized our production controls into an enterprise kit.
-
Open source
ethos-gate
A PreToolUse safety gate that blocks irreversible or outward-facing agent actions unless a human approved that exact content.
Role: I built the gate and extracted it from the system I run.
-
Open source
warden
A governor for autonomous agents that admits autonomy by reversibility rather than confidence, with signal-bound rollback and offline tests.
Role: I designed and built it, with four production post-mortems behind it.
-
Open source
orchestra-agents
Multi-agent orchestration on Claude with bounded fan-out, a durable message bus, circuit breakers, best-of-N with an adversarial judge and permission tiers.
Role: I extracted the production orchestration core and wrote the eval harness.
-
Live product
BYU AI Foundry
An AI product studio and experiential learning program of BYU Marriott where student builders ship production AI for real clients.
Role: Founder and Managing Director
BYU Marriott School of Business
-
Open source
claude-bug-squash
Triages a repo, fixes in an isolated worktree, proves red to green, and gates every fix behind blast-radius caps and a three-seat review.
Role: I built it as a gated autonomous bug fixer that can never merge.
-
Open source
Claude Code behavior specs
Target-behavior specs for the Claude Code CLI with observed versus expected behavior, runnable reproductions and proposed evals.
Role: I wrote them from production incidents.
-
Open source
agent-evals-starter
Learn agent evals by running them: a support agent, 24 tasks, code graders, an LLM judge and do-nothing controls, all offline.
Role: I built it as a teaching repo for agent evals.
-
Open source
ccstitch
A local trace store for Claude Code OpenTelemetry spans that repairs orphaned headless turns and nests child runs under the launching tool call.
Role: I built it to fix a gap I hit running headless Claude Code.
-
Open source
Clawdling
Shows your Claude Code projects as an org on one map, who is working and who needs you, with a receipt when an agent says done.
Role: I built it as an open-source map of a person's Claude Code agents.
-
Open source
deep-research-agent
Iterative multi-source research with a knowledge-gap loop and per-claim refute-or-survive verification.
Role: I built it; the per-claim verify stage is ported, with credit, from Claude Code's /deep-research skill.
-
Open source
ai-phone-desk
A Vapi voice agent where nothing dials until a human approves the exact request.
Role: I built it as a human-approval pattern for voice agents.
-
Open source
clawd-agent-os
Orchestrates Claude agents in tiered hierarchies with worktree-isolated fleet spawning and a durable bus that never silently returns empty.
Role: An earlier extraction of the system I run. orchestra-agents is the production core.
-
Open source
mastery-engine
An evidence-grounded curriculum generator with mastery gates where claims are verified against source material on disk.
Role: I built it.
-
Writing
Field Notes docs
Plain-language docs on how to build and run a production AI organization with Claude Code, plus AI Core Skills.
Role: I wrote the docs.
-
Live product
Clawdacademy
A site that shows how well you use AI and where to improve.
Role: I built the interactive AI fluency trainer.
-
Live product
Vibe Coding Detector
Scores any public Git repo for AI-pattern signals and maintainability with deterministic heuristics. It flags patterns; it does not prove AI authorship.
Role: I built it.
-
Live product
BookJD
A booking app on Next.js, Supabase and Google Calendar availability, live and open source.
Role: I built it as a self-hosted Calendly replacement.
Family venture: go-to-market and SEO
Products built with Fable Software, a family venture, where my work was mostly go-to-market, SEO, positioning and build audits.
-
Live product
SpeedPrep
A math speed game for kids that teaches 84 shortcuts, with timed tests under 5 minutes and races with friends on nearby phones.
Role: Built the Speed Mode V2 arena and led the two-mode restructure, the App Store and Play launch kit, a UX audit and the launch ad plan.
with Fable Software
-
Live product
SolarKnock
A door-to-door canvassing app for solar sales teams, with territories, leaderboards, instant solar estimates and homeowner data.
Role: Go-to-market, SEO and positioning, plus a build audit.
with Fable Software
-
Live product
ReadyStack
Free Docker images of open-source Community Edition software, with handbooks checked against the exact image, tested seed data and a build kit that rebuilds it offline.
Role: Product requirements and workplan, SEO and developer-adoption plan, plus a build audit.
with Fable Software
-
Live product
BeefShopper
A ranking site that compares 50+ home-delivery ribeyes from 28 sellers by price per pound, USDA grade, aging and sourcing, with every claim cited.
Role: Go-to-market, SEO and affiliate-growth plan, plus a build audit and market research.
with Fable Software
-
Live product
TrendZion
An illustrated atlas of Latter-day Saint words and ideas where people compare how teachings and language changed across nearly two centuries of sermons and scripture.
Role: Positioning and rebrand plan, plus a build audit and market research.
with Fable Software
-
Live product
Deckboard
A free web app that turns any screen with a browser into a photorealistic split-flap display, with engineered sound, live channels and a documented API.
Role: SEO and go-to-market plan, plus a build audit and competitor research.
with Fable Software
Nothing matches that search. Try a broader word, or clear the filters.