Everything we've built.
My own builds, the BYU AI Foundry's work, Spark AI at Siemens, and paid client work. Every card links to the real thing.
78 projects
Start with these
Our top ten, picked so a stranger can click every one.
- 01Spark AI at Siemens USA 120+ agent AI platform for 1,000+ Siemens US employees; adoption went from 3% to 65% in 10 weeks.
- 02BYU AI Foundry (the program)An AI-native product studio and consultancy at BYU Marriott where students build production AI software for real clients.
- 03AI Foundry MarketplaceFree MIT-licensed plugin marketplace for Claude Code: an 8-stage SDLC plus skills, agents and hooks, each labeled by how well tested.
- 04App Feedback LoopOpen build spec for an in-app feedback button: one click captures a screenshot and context, an agent fixes it, the fix commit thanks the reporter.
- 05Nerve Center: the AI organization JD runs his work onA self-hosted AI operating system: one orchestrator over 30+ agents in daily production, with human gates on anything irreversible.
- 06recruit-copilotA job search as a verification problem: experience bank, role scoring, resumes proven machine-readable, no auto-apply by design.
- 07Clawdling: Claude Code projects as one org mapYour Claude Code as a whole org: an agent for each project on one map, showing who is working and who needs you, with a receipt when one says done.
- 08SpeedPrepACT and SAT math prep where an algorithm picks the next lesson, quiz or review for each student.
- 09DeckboardTurn any screen into a photoreal split-flap (airport departures board) display, free.
- 10AgentTree Merch: an AI-operated storeA print-on-demand t-shirt store run by an agent loop, with money, identity and IP sign-off kept human.
The library
Showing all 78 projects
-
AI Foundry · Plugin
AI Foundry Marketplace
Live
Free MIT-licensed plugin marketplace for Claude Code: an 8-stage SDLC plus skills, agents and hooks, each labeled by how well tested.
Details and links: AI Foundry Marketplace
Install with one /plugin command into your own Claude Code. The catalog index lists 53 items (32 skills, 12 agents, 8 hooks) across 9 plugins: foundry-sdlc, guardrails, qa, design, devices, iphone-mirror, evals, builder-kit and recruit. Each item carries an honest label: 32 verified (passed a live end-to-end run), 18 beta and 3 new. The framework encodes norms such as root-cause-first, never say done without a passing gate, and review by blast radius. A browsable site and machine-readable catalog-index.json let an agent pick the smallest set of plugins for a project.
-
AI Foundry · Program
BYU AI Foundry (the program)
Live
An AI-native product studio and consultancy at BYU Marriott where students build production AI software for real clients.
Details and links: BYU AI Foundry (the program)
AI Foundry is an official program of BYU Marriott, administered by BYU Strategy, with Scott Murff as faculty advisor and JD Davenport as Managing Director. The 2026-27 cohort is 30 students (15 MBAs, 15 undergrads) working in teams of one or two on five-week client engagements. Students first learn the Builder toolkit, then work as builder consultants in the Foundry Lab (MBA 693R / STRAT 490R). AI Foundry is a member of Anthropic's Claude Partner Network, and the public roadmap tracks the network, advisors and certifications as they grow.
- Official site (aifoundry.byu.edu)
- Our approach (aifoundry.byu.edu)
- For students (aifoundry.byu.edu)
- Oct 2026 update (aifoundry.byu.edu)
- Org on GitHub (github.com)
-
JD · Harness
Nerve Center: the AI organization JD runs his work on
Live
A self-hosted AI operating system: one orchestrator over 30+ agents in daily production, with human gates on anything irreversible.
Details and links: Nerve Center: the AI organization JD runs his work on
Nerve Center is the private multi-agent system JD designed, built and operates on a Mac mini. The public showcase page (snapshot 2026-11-09) reports 30+ agents in daily production, 100+ built over time, 310 background jobs (65 of them watchdogs) and 1,092 shipped-work log entries in 30 days. Work arrives by Telegram voice/text or a web and iOS cockpit, goes through an orchestrator and a model router, becomes a durable job-queue row, and anything outward-facing or irreversible stops at a gate until JD says yes. The build itself stays private because it runs his life and work; the public teardown covers architecture, three production failures with root causes, and cost.
- Live showcase (jddavenport.com)
- Live brain (counts live) (jddavenport.com)
- Production teardown (jddavenport.com)
Visit site: Nerve Center: the AI organization JD runs his work on
-
JD at Siemens · Case study
Spark AI at Siemens US
Shipped
A 120+ agent AI platform for 1,000+ Siemens US employees; adoption went from 3% to 65% in 10 weeks.
Details and links: Spark AI at Siemens US
In summer 2026 JD worked at Siemens as an AI Product Strategist in the XPS Program, focused on the Siemens US organization. He built and launched Spark AI, a custom platform of 120+ agents and 50+ skills built for specific job functions and seniority levels, deployed to 1,000+ employees. Active adoption rose from a 3% baseline to 65% in 10 weeks, with 650+ employees active by summer's end. He also wrote and presented a 3/6/12-month AI roadmap to US leadership, sequenced by function readiness with role-specific success metrics. His manager originated the idea; JD built, launched, drove adoption and wrote the roadmap. He wrote a product spec first and built to checklists with acceptance criteria. Lesson he names: more QA before each release and smaller, steadier releases.
-
JD · Course
agent-evals-starter
Live
Learn agent evals by running them: a support agent, 24 tasks, code graders, an LLM judge and do-nothing controls, all offline.
Details and links: agent-evals-starter
MIT teaching repo that includes a .claude/skills/agent-evals skill and two agents (eval-task-auditor, judge-writer). Accompanies the AI Core Skills #1 agent-evals series on the docs site.
-
JD · Tool
claude-bug-squash
Live
A gated autonomous bug fixer: triage, a fix in an isolated worktree, red-to-green proof, and a three-seat review before anything merges.
Details and links: claude-bug-squash
Point Claude Code at a repo and report a bug. It triages and assigns a risk tier, writes a fix in a throwaway git worktree, proves the bug goes red to green, then runs the real diff through blast-radius caps, a never-touch deny-list and a 3-seat adversarial review panel. Merging is off by default; the user reads proposed diffs first. Public MIT plugin and standalone CLI.
-
JD · Tool
Clawdling: Claude Code projects as one org map
Live
Your Claude Code as a whole org: an agent for each project on one map, showing who is working and who needs you, with a receipt when one says done.
Details and links: Clawdling: Claude Code projects as one org map
Clawdling is the open-source, bring-your-own-key extraction of the idea behind JD's private system. It runs locally on your own Claude plan, gives each project an agent, and puts them on one map showing who is working and who needs you. A receipt is shown when an agent claims it is done. It is AGPL-3.0 and is the engine under Clawdascended, a hosted product JD says is still early. The build write-up covers the bugs that only appeared when it was tested as a stranger from a fresh clone.
-
JD · Tool
ethos-gate
Live
A PreToolUse safety gate that blocks irreversible or outward-facing agent actions unless a human approved that exact content.
Details and links: ethos-gate
A MIT-licensed PreToolUse gate for autonomous agents with 4 hooks, 62 tests and a lethal-trifecta auditor, extracted from JD's own estate. Installable as a Python package; no plugin.json found in the tree, so it is not one-command plugin-installable.
-
AI Foundry · Plugin
Foundry Evals plugin
Live
Scaffold an eval suite with oracles, code graders, a binary judge and negative controls; run, compare and hill-climb it.
Details and links: Foundry Evals plugin
Skills new-eval, run-eval, hillclimb and agent eval-auditor. Evidence: 3 verified, 1 beta. Pairs with the AI Core Skills agent-evals docs. Part of the foundry-marketplace; MIT.
-
AI Foundry · Plugin
Foundry Guardrails plugin (hooks)
Live
Hooks that block destructive commands and secret leaks and require tests before done, push or deploy.
Details and links: Foundry Guardrails plugin (hooks)
Hook-only plugin plus a guardrails-help skill. Described by its authors as a seatbelt, not a vault; limits are listed. Evidence: 6 verified, 1 beta. Part of the foundry-marketplace; MIT.
-
AI Foundry · Plugin
Foundry SDLC plugin
Live
The 8-stage 0-to-1 workflow: scaffold, requirements, PRD and stories, workplan, parallel lanes, QA, ship, feedback.
Details and links: Foundry SDLC plugin
8 skills (new-project, gather-requirements, write-prd, write-workplan, parallel-lanes, ship, bug-to-story, sdlc) plus 2 judge agents (prd-critic, story-verifier) and hooks. Evidence: 10 verified, 1 beta. Part of the foundry-marketplace; MIT.
-
JD · Website
JD Davenport docs: Builds, AI Core Skills, Field Notes
Live
Public docs site with build write-ups, AI Core Skills (agent evals first), field notes and a production teardown.
Details and links: JD Davenport docs: Builds, AI Core Skills, Field Notes
Already live in jddavenport-site-public/docs with a docs/search-index.json (113 entries). Sections: AI Core Skills 9, Start here 10, Fundamentals 13, Patterns 16, Inside the Nerve Center 20, Safety and operations 6, Builds 19, Field notes 3, Product management 3, Reference 4, plus section pages. Includes an open-source index (reference/open-source) and build-your-own guides (resume grader, merch store, first agent).
- Docs (jddavenport.com)
- Builds index (jddavenport.com)
- Work (jddavenport.com)
- Open-source index (jddavenport.com)
- Site repo (github.com)
Read the write-up: JD Davenport docs: Builds, AI Core Skills, Field Notes
-
JD · Tool
recruit-copilot
Live
A job search as a verification problem: experience bank, role scoring, resumes proven machine-readable, no auto-apply by design.
Details and links: recruit-copilot
Public MIT repo listed as the external 'recruit' plugin in the Foundry marketplace. Skills resume-intake, resume-builder, resume-grader plus three grader agents (recruiter, hiring manager, AI-systems rep). It deliberately does not auto-submit applications. Needs Python 3.9+ and nothing to pip install. Marketplace evidence label: beta.
-
JD · Tool
Resume grader (powers the Foundry jobs board)
Live
One Claude call scores any resume against any job on a six-dimension rubric and returns specific feedback.
Details and links: Resume grader (powers the Foundry jobs board)
Built July 2026 for the BYU AI Foundry jobs board, which launched with about 10,000 openings. It is an LLM-as-judge, explicitly not RAG: the rubric is universal and the posting is pasted into the prompt. It was the only AI running inside the product; the scraper and site were ordinary code. Open-sourced under MIT. A build-your-own tutorial shows a 40-line Python version.
-
AI Foundry · Tool
Agent Team Kit
Live
A working setup for running many Claude Code agents on one app without collisions, from mockup to release to bug reports.
Details and links: Agent Team Kit
Started as a fix for parallel agents colliding on ports, the database and the git tree. Two modes (stay on main, or per-agent git worktrees with a cockpit session that merges) share the same hooks: dev-port assignment, schema lock, git guard, file-size ratchet, import boundaries, pre-commit and pre-push gates. It also ships a /mockup skill for clickable prototypes, a memory pack, /fix-bug and /ux-critique skills, specs for in-app feedback loops, and notes on tending agents from a phone. The repo has no license file yet, so ask before you reuse it.
-
JD · Business
AgentTree Merch: an AI-operated store
Live
A print-on-demand t-shirt store run by an agent loop, with money, identity and IP sign-off kept human.
Details and links: AgentTree Merch: an AI-operated store
shop.agenttree.army designs, screens for trademark risk, generates art, lists products, takes Stripe payment and sends paid orders to Printful. The write-up is blunt: as of mid-June 2026 it had two orders, both JD's own, and it is an engineering case study, not a revenue story. First real order found three fulfillment bugs that a synthetic harness missed, plus a fourth that charged without printing. There is also a step-by-step guide to build your own and an IP-safety-gate article on why a classifier must fail closed.
- Store (shop.agenttree.army)
- Build write-up (jddavenport.com)
- First-order bugs (jddavenport.com)
- Build your own (jddavenport.com)
- IP-safety gate (jddavenport.com)
-
AI Foundry · Website
aifoundry.byu.edu official website
Live
The program's public site: work, approach, student path, network signup, project requests, roadmap and donors.
Details and links: aifoundry.byu.edu official website
A Next.js and Tailwind site in the public repo ai-foundry-byu/website, deployed on Vercel under BYU's domain. Pages cover Our work, Our approach, For students, About, Join the network, Start a project, the Roadmap, Donors and a monthly Update. The roadmap page reads its counts live from the program's own database.
-
AI Foundry · Tool
App Feedback Loop
Live
Open build spec for an in-app feedback button: one click captures a screenshot and context, an agent fixes it, the fix commit thanks the reporter.
Details and links: App Feedback Loop
Not a library: a SPEC.md that a coding agent implements natively in your stack. Install is a single prompt telling Claude Code to read the spec and run the checks in VERIFY.md. The loop covers a widget (also Cmd/Ctrl+Shift+F), a context bundle (route, breadcrumbs, console, network, errors, environment, screenshot) with client-side secret scrubbing, an intake endpoint, an admin inbox, a /fix-bug skill, and a git hook that closes the loop to the reporter. Featured first on the Our work page and built into every app the Foundry ships.
-
AI Foundry · Case study
CareBridge Connect
Live
AI-native system that turns EHR staff notes into plain-language updates for approved family members at nursing facilities.
Details and links: CareBridge Connect
Listed on the public Our work page as a shipped, running product for skilled nursing and acute care. It converts the notes staff already write into updates on changes of condition, appointments and events, with role-based access, encryption and audit logging on every PHI access. The page says it is onboarding its first five facilities, each over 100 residents.
-
JD · Harness
clawd-agent-os
Live
Orchestrate 50+ Claude agents: worktree fleets, a durable agent bus, single-registry model tiering.
Details and links: clawd-agent-os
Public MIT repo on jddavenportOpen (descriptions from gh repo list). Library/CLI style code repo; no SKILL.md, plugin.json or agent .md found in the tree, so offer as 'clone and run', not plugin install. Read the README before writing detailed claims.
-
JD · Agent
Course-expert agents
Live
One agent per enrolled course that answers from the real slides and syllabus, cites the exact slide, and knows what is due.
Details and links: Course-expert agents
Started April 2026 with one analytics course, then one expert per course from fall 2026, plus a seat for a course JD TAs. Retrieval is built offline; deadlines come from the LMS. The write-up lists five failures. Same-question disagreement between three general models on 2026-04-21 motivated it.
-
JD · Agent
deep-research-agent
Live
Iterative multi-source research agent: knowledge-gap loop, devil's-advocate pass, per-claim refute-or-survive verification.
Details and links: deep-research-agent
Keeps searching until it cannot name a gap, argues against itself, then tries to refute every factual claim. The per-claim verify stage is ported with credit from Claude Code's /deep-research skill. The build page also covers two embarrassing failures of the agent.
-
JD · Research
Deploying a 30-agent AI organization safely
Live
A capability-versus-risk case study of running about 38 live agents with real consequences.
Details and links: Deploying a 30-agent AI organization safely
Written 2026. For each capability (acting on a live machine, outbound communication as a real person, and others) it lists the risk, the controls and the tradeoffs. Explicitly scoped as a single-operator system, not a multi-tenant solution.
-
AI Foundry · Plugin
Foundry Builder Kit plugin
Live
Day-one build skills: add auth, 10k-user launch audit, secrets hygiene, cheap LLM routing, scheduled agents, tmux agents.
Details and links: Foundry Builder Kit plugin
6 skills with runnable scripts and templates (launchd, cron, GitHub Actions, pre-commit secret scan) and tests. Evidence: 4 verified, 3 beta; generated code is unit-tested but not run end to end, per the README. Part of the foundry-marketplace; MIT.
-
AI Foundry · Plugin
Foundry QA plugin
Live
Definition-of-done gate: deployed SHA plus a Playwright journey on desktop, phone and tablet, regression and a11y checks.
Details and links: Foundry QA plugin
6 skills (done-gate, regression-check, a11y-check, explore, persona-run, visual-baseline) and 3 agents (web-qa, explorer, persona-tester). Evidence: 6 verified, 3 beta. Part of the foundry-marketplace; MIT.
-
JD · Case study
National Grid: cloud and AI platform
Archived
Principal PM who grew an internal cloud platform from $2M to $20M+ chargeback and launched a RAG GenAI platform.
Details and links: National Grid: cloud and AI platform
2022-2025 at a 20,000-person utility, promoted Technical to Senior to Principal PM, owning the cloud and AI/ML roadmap. Launched a RAG-based internal GenAI platform (LLM over proprietary documentation) reaching 3,000+ monthly users. Built an IaC service standardizing 100+ enterprise tools across business units; deployment went from weeks to minutes. Founded Cloud Leadership Sync, the first executive and Chief Architect cross-BU forum. Grew team from 1 to 4 pods and hired 4 PMs; engineering org grew 5 to 35.
-
JD · Tool
Plaud voice-note pipeline and openplaud
Live
Recorder to local transcript to tasks, CRM notes and memory, plus an open-source self-hosted Plaud manager.
Details and links: Plaud voice-note pipeline and openplaud
JD carries a Plaud recorder. Recordings sync locally, are transcribed with Whisper, then action items go to task lists, people to a CRM and context to notes. Version one was confidently wrong and polluted memory, so most of the write-up is about precision. openplaud is a free self-hosted recording manager with Whisper, Groq or Deepgram, search and diarization, needing no Plaud subscription.
-
JD · Newsletter
Raising Clawd newsletter and pipeline
Live
JD's newsletter on what holds up, breaks and costs when building an AI organization, with an approval-gated weekly drafter.
Details and links: Raising Clawd newsletter and pipeline
A weekly drafter builds an evidence pack from the last seven days of changelogs and git history, writes the issue in a fixed structure, and runs six guards on it before a preview goes to JD. Nothing sends without his approval, and the approval is bound to the exact content, so a one-byte change voids it. Sends go only to confirmed subscribers, with one-click unsubscribe.
-
JD · Agent
ai-phone-desk
Live
A Vapi voice agent where nothing dials until a human approves the exact request.
Details and links: ai-phone-desk
A human-approval pattern for voice agents: nothing dials until a person approves the exact request. Built on a Vapi voice agent with content-bound approval, for your own agent stack or an outside agent.
-
JD · Tool
Analytics suite (30/30 on practice final)
Built
Seven statistical playbooks, a classifier, a solver and a verifier; tool use beat model choice.
Details and links: Analytics suite (30/30 on practice final)
Built in a day in April 2026 for a take-home MBA analytics final where the course allowed AI. The setup that computed its answers scored 30 of 30 on the practice exam; models answering cold landed around 15 to 21 regardless of which model it was. Tool use beat model choice.
Read the write-up: Analytics suite (30/30 on practice final)
-
JD · Tool
claude-deploy-kit
Live
Enterprise controls for deploying Claude agents: policy chokepoint, destructive-command deny hook, eval gate, redacting audit log.
Details and links: claude-deploy-kit
Public MIT repo on jddavenportOpen (descriptions from gh repo list). Library/CLI style code repo; no SKILL.md, plugin.json or agent .md found in the tree, so offer as 'clone and run', not plugin install. Read the README before writing detailed claims.
-
JD · App
Clawdacademy (AI fluency trainer)
Live
Scores seven judgment behaviors in your real Claude Code sessions, names your weakest habit and coaches in-flow.
Details and links: Clawdacademy (AI fluency trainer)
Local-first, free, MIT. Cites the METR study (developers felt 20% faster but measured 19% slower). Live at clawdacademy.app; repo is ai-fluency-trainer.
-
JD · Skill
contact-list-builder skill
Live
A Claude Code skill that builds verified public contact lists; a script re-downloads every source page to prove each email and phone.
Details and links: contact-list-builder skill
Six to eight parallel research subagents split the search by segment and region. A script, not the model, then re-downloads every source page and checks that each email and phone is really there. First real run: 213 contacts across 6 segments in about 15 minutes, every one confirmed against its source page.
-
JD · Tool
Context Kit
Live
Four Markdown templates and five Claude Code skills that give an AI permanent personal context; installs in 60 seconds.
Details and links: Context Kit
MIT repo with templates for a personal wiki, mental models, voice and protocols files and skills to use them. Install is a curl-to-bash script.
-
AI Foundry · Plugin
Foundry Design plugin
Live
A subtract-only design critic, UI spec writer with full state matrix, and design-system template.
Details and links: Foundry Design plugin
Skills design-critique, ui-ux-spec, design-system and agent ui-reviewer. Evidence: 2 verified, 2 beta. Part of the foundry-marketplace; MIT.
-
JD · Tool
mcp-judge
Live
A calibrated multi-persona LLM judge exposed as an MCP server, shipped with an eval harness that proves the calibration.
Details and links: mcp-judge
Public MIT repo on jddavenportOpen (descriptions from gh repo list). Library/CLI style code repo; no SKILL.md, plugin.json or agent .md found in the tree, so offer as 'clone and run', not plugin install. Read the README before writing detailed claims.
-
JD · App
OpenBudget
Live
Self-hosted envelope-budgeting app with read-only bank sync, about $15 a year versus a roughly $109 subscription.
Details and links: OpenBudget
Built for JD first, scrubbed and published June 2026 under MIT. Bank data via SimpleFIN Bridge into the user's own Supabase; Next.js app with magic-link sign-in.
-
JD · Tool
orchestra-agents
Live
Production multi-agent orchestration on Claude: bounded fan-out, durable message bus, circuit breakers, best-of-N with an adversarial judge.
Details and links: orchestra-agents
Public MIT repo on jddavenportOpen (descriptions from gh repo list). Library/CLI style code repo; no SKILL.md, plugin.json or agent .md found in the tree, so offer as 'clone and run', not plugin install. Read the README before writing detailed claims.
-
JD · Tool
Warden: autonomy admitted by reversibility
Live
A governor for autonomous agents that admits autonomy by reversibility, not confidence, with signal-bound rollback and offline tests.
Details and links: Warden: autonomy admitted by reversibility
Warden is an open-source governor distilled from four production post-mortems on JD's own agent estate. The principle is that an agent earns autonomy for actions that can be rolled back, not for actions it feels confident about. It ships signal-bound rollback and offline tests.
-
JD · Tool
AI decks that don't look like AI
Live
Eight failed python-pptx attempts, then a design system plus a render-read-critique-fix loop.
Details and links: AI decks that don't look like AI
From April to June 2026 JD learned that layout decisions must leave the model and the model must look at the rendered slide. This is how his decks are built today.
-
JD · App
CasePrep and caseprep CLI
Built
Paste an MBA case, get frameworks, key questions and analysis in about 60 seconds.
Details and links: CasePrep and caseprep CLI
Next.js 14 with Better Auth, Turso, Stripe subscriptions and GPT-4o-mini. A separate local CLI extracts financials and entities from case PDFs and builds a prep template.
-
JD · Tool
ccstitch
Live
Local trace store for Claude Code OpenTelemetry spans that repairs orphaned headless turns.
Details and links: ccstitch
Public MIT repo on jddavenportOpen (descriptions from gh repo list). Library/CLI style code repo; no SKILL.md, plugin.json or agent .md found in the tree, so offer as 'clone and run', not plugin install. Read the README before writing detailed claims.
-
AI Foundry · Milestone
Claude Partner Network and Claude Architect certification
Live
AI Foundry is a member of Anthropic's Claude Partner Network, and its cohort leaders are working through the Claude Certified Architect exam.
Details and links: Claude Partner Network and Claude Architect certification
Members complete four Anthropic courses (Agent Skills, Building with the Claude API, Model Context Protocol, and Claude Code in Action), then sit the Claude Certified Architect exam, with group study sessions along the way. The public roadmap shows how many cohort leaders have passed.
Visit site: Claude Partner Network and Claude Architect certification
-
AI Foundry · Course
Foundry Lab student program guide
Live
Quarto site with the course syllabus, certification track, staffing model, client update norms and build standards.
Details and links: Foundry Lab student program guide
The course site for the Foundry Lab. It covers the AI Foundry Constitution and Operating Model, the Client Service Selection Policy, The Builder and the Consultant, the syllabus and sprints, the Claude certification track with a practice exam, and how client teams are staffed.
-
AI Foundry · App
French 80/20
Live
AI-native French learning app for web and iPhone focused on the words you actually use and progress toward 80% of everyday French.
Details and links: French 80/20
Shown on Our work as a reference build that is shipped and running on web and iPhone. The public description stresses personalized vocabulary and measured progress toward 80% comprehension of everyday French.
-
JD · Tool
harnessview
Live
Scans a Claude Code project's agents, skills, hooks, MCP and settings and draws one live graph with broken wiring in red.
Details and links: harnessview
Public MIT repo on jddavenportOpen (descriptions from gh repo list). Library/CLI style code repo; no SKILL.md, plugin.json or agent .md found in the tree, so offer as 'clone and run', not plugin install. Read the README before writing detailed claims.
-
AI Foundry · App
LawGrader / WriteGrader
Live
Agentic grading for legal and academic writing: it drafts scores and margin comments, the instructor decides every final grade.
Details and links: LawGrader / WriteGrader
Two live sites for agentic grading of legal and academic writing. The human-in-the-loop design is the point: AI drafts scores and margin comments, and the instructor decides every final grade.
-
JD · App
BookJD
Live
Self-hosted Calendly replacement on Next.js, Supabase, Google Calendar free/busy and Stripe.
Details and links: BookJD
Live and open source, powers the booking flow on JD's site.
-
JD · Case study
Brio Energy: sales and accounting systems
Archived
Built a custom CRM that cut sales-to-accounting turnaround from 5 days to 15 minutes.
Details and links: Brio Energy: sales and accounting systems
2018-2020 Senior Full-Stack Developer and Product Owner at a solar company. Built a CRM around the sales-to-accounting handoff, a lead-generation site producing 175+ qualified leads a week (contributing to $3.9M annual new revenue), and microservices that cut accounting spend by $745K a year against prior-year run rate.
Read the write-up: Brio Energy: sales and accounting systems
-
JD · Research
Claude Code behavior specs
Live
Target-behavior specs for the Claude Code CLI written from production incidents, with runnable reproductions and proposed evals.
Details and links: Claude Code behavior specs
Observed-versus-expected behavior of Claude Code, written from incidents in a roughly 55-agent estate.
-
JD · Harness
Fleetwright
Built
Self-hosted agent operations OS: a CEO orchestrator, domain agents, a Next.js cockpit and a fleet engine in git worktrees.
Details and links: Fleetwright
Public sibling of the private Nerve Center; FastAPI bridge plus Next.js cockpit.
-
AI Foundry · Plugin
Foundry Devices plugin
Beta
Native QA agents for iOS Simulator, Android, Windows and macOS plus a device-matrix skill.
Details and links: Foundry Devices plugin
4 agents (android-qa, ios-qa, macos-qa, windows-qa) and 4 skills. Evidence: mostly beta, 1 new; passes on CI runners but not yet on a student's own app per README. Part of the foundry-marketplace; MIT.
-
JD · Agent
LinkedIn and content pipeline
Live
Idea sourcing, drafting in JD's voice, a second-model review and a per-post approval card; one door to publish.
Details and links: LinkedIn and content pipeline
Approval is void if one word changes after yes. A comment drafter never posts on its own. The write-up covers a bug that hid its drafts for weeks.
-
JD · Tool
mastery-engine
Built
Evidence-grounded curriculum generator; every claim is verified against code on disk before you see it.
Details and links: mastery-engine
Public MIT repo on jddavenportOpen (descriptions from gh repo list). Library/CLI style code repo; no SKILL.md, plugin.json or agent .md found in the tree, so offer as 'clone and run', not plugin install. Read the README before writing detailed claims.
-
JD · App
PitchGrade
Built
Upload a pitch deck PDF and get scored on 12 VC criteria with feedback.
Details and links: PitchGrade
A pitch-deck grader built with Next.js 14 and GPT-4o-mini, with a Stripe plan.
-
JD · Agent
Socrates: the X autopilot
Archived
A self-learning X content agent that was paused; the lesson is safety rails arriving after the architecture.
Details and links: Socrates: the X autopilot
Planned content, drafted posts in several voice genomes, tested variants and posted on a schedule on a separate persona. As of October 2026 its loops are off and nothing posts automatically. JD says he learned more from how it failed than from any other build.
-
AI Foundry · Research
Software Design Canon
Live
A link-checked index of the genuinely free primary sources on software design and architecture.
Details and links: Software Design Canon
Points to the original papers, articles, catalogs and talks behind famous design books (SOLID, hexagonal architecture, modularity and more) plus free books. It does not host pirated text and has a per-book page on what is free versus must be bought. A GitHub Action re-checks every link weekly and opens an issue on a dead one. Reading paths and a machine-readable sources.yml let an agent cite a primary source instead of a paraphrase.
-
JD · App
VentureVal
Built
Startup valuation calculator using five VC/PE methods (Berkus and others) with a comparison dashboard.
Details and links: VentureVal
Built for an MBA venture capital and private equity class.
-
JD · Tool
Vibe Coding Detector
Live
Scores a public Git repo for AI-pattern signals and maintainability with deterministic heuristics.
Details and links: Vibe Coding Detector
Flags patterns; it does not prove AI authorship, and the README says so up front.
-
JD · Agent
VoiceClaw
Archived
A phone number you call to talk to an AI wired to calendar, email, messages and notes. Built March 2026, now archived.
Details and links: VoiceClaw
Vapi handled telephony, Deepgram transcription, Claude did tool calls via a small FastAPI tools server. The build page explains why it stopped, which is useful for any team bolting on a second channel.
-
AI Foundry · Plugin
Foundry iPhone Mirror plugin
Beta
Drive a real iPhone from a Mac through iPhone Mirroring: screenshot, read, tap, type, swipe.
Details and links: Foundry iPhone Mirror plugin
Skill, iphone-driver agent and an MCP server. Labeled new (experimental); README says it has never driven a real phone in tests. Part of the foundry-marketplace; MIT.
-
JD · Skill
instinct-standing-doc (handoff skill + AGENTS.md)
Live
Routing rules, a Claude Code skill and an AGENTS.md for handing work to the Instinct autonomous agent. Unofficial.
Details and links: instinct-standing-doc (handoff skill + AGENTS.md)
An MIT repo with a routing doc, a Claude Code skill and an AGENTS.md for handing work from your agent system to Instinct, an outside autonomous agent. Unofficial, and useful only if you use Instinct.
View repo: instinct-standing-doc (handoff skill + AGENTS.md)
-
AI Foundry · App
Project ingestion portal (Submit a Project)
Built
Intake for organizations to propose work, with a review queue; the front door that became the site's Start a project page.
Details and links: Project ingestion portal (Submit a Project)
PRD and workplan from June 2026 describe a public proposal form, database of proposals and admin approve/decline queue at a flat fee. The shipped version lives on aifoundry.byu.edu/quote.
-
JD · Tool
AGI readiness auditor and AI spend tracker
Built
A CLI that pressure-tests public AGI-safety claims, and a tracker for AI lab spend.
Details and links: AGI readiness auditor and AI spend tracker
Two small research tools from the same line of work.
-
JD · Tool
Crypto analytics engine (educational)
Built
Coin-agnostic educational confluence-analytics engine: TA, derivatives, on-chain, macro, sentiment.
Details and links: Crypto analytics engine (educational)
An educational analytics engine that combines technical analysis, derivatives, on-chain, macro and sentiment signals for any coin. Educational only. Not financial advice.
-
JD · App
OpenRemarkable
Built
Free self-hosted reMarkable manager with local handwriting OCR, full-text search and export.
Details and links: OpenRemarkable
Replaces a $3 to $8 monthly cloud subscription; REST API included.
-
JD · Tool
BYU Outlook browser integration
Built
Browser automation for BYU Outlook and Teams where no API exists.
Details and links: BYU Outlook browser integration
Public repo of the technique; the live version in JD's system signs in with stored credentials.
Client work
Paid consulting engagements, built end to end with the client: product, engineering and go-to-market.
-
Client work · Website
ce.rodeo
Live
A free daily index that grades the Docker images of more than 150 open-source projects by security and health.
Details and links: ce.rodeo
ce.rodeo answers 'which Docker image should I pull?' by lining up every publisher's image for more than 150 open-source projects, scanning each for known vulnerabilities, and giving a letter grade from a published formula. It ranks publishers, lists images whose grade dropped this week, and has a free public data API. It began scanning in October 2026.
-
Client work · App
SpeedPrep
Live
ACT and SAT math prep where an algorithm picks the next lesson, quiz or review for each student.
Details and links: SpeedPrep
SpeedPrep is a web app (with iOS and Android builds in progress) that teaches ACT and SAT math using mastery learning, spaced repetition and interleaved review, enforced by an algorithm rather than student discipline. Every question is timed, and a projected ACT score updates as the learner practices. A free Prep Assessment gives an instant estimate. One parent account covers up to 10 learners. It is tested with a pixel-level visual test suite and a synthetic-student test that plays every game mechanic. Beta is free; math comes first, English, Reading and Science are planned. Includes a 820-question bank checked with an outside classifier (about 91% top-1 topic agreement).
-
Client work · App
Deckboard
Live
Turn any screen into a photoreal split-flap (airport departures board) display, free.
Details and links: Deckboard
Deckboard runs in the browser on any TV, tablet or monitor and shows a Solari-style split-flap board with hinge shadows, flap motion and sound modelled on the real machine. It has live channels and a documented push API so developers can send data from cron jobs or webhooks. Positioned as free with no tiers, against hardware boards and subscription services. Aimed at cafes, offices, venues and makers.
-
Client work · Tool
ReadyStack
Live
Clean-room, offline-rebuildable Docker images of open-source Community Editions, with provenance.
Details and links: ReadyStack
ReadyStack publishes verified Docker images of open-source Community Edition software, each built with the network disabled so it has no hidden dependencies, with SHA256 stamps, SBOM and signed attestations. Images are free on DockerHub; the paid product is a full archive (source, dependencies, docs, tests) for audit and air-gapped use. Built for AI agents and enterprise auditors who need to verify what they pull.
-
Client work · App
SolarKnock
Live
A canvassing app for door-to-door solar sales: territories, leaderboards and instant solar estimates.
Details and links: SolarKnock
SolarKnock replaces the mix of mapping tool, spreadsheet, CRM and group chat that solar door-knocking teams use. Reps get live territory pins, quick power-bill and solar savings estimates on the doorstep and competitive leaderboards; managers get a dashboard and dispatch. It runs as a marketing site plus an app at app.solarknock.com, shipped as an installable web app built with Expo / React Native. Currently a free beta; paid team tiers are shown on the site. Operated under Fable Software LLC. The App Store launch and text-message (A2P) registration are still open work.
-
Client work · Research
SpeedPrep Arcade (ten game prototypes)
Built
Ten playable prototypes testing whether a game layer could make math practice stick.
Details and links: SpeedPrep Arcade (ten game prototypes)
A static site of ten feel-prototypes for a game layer on SpeedPrep, all sharing one 56-item ACT question pool so the games can be compared fairly. Concepts were picked by a games-expert agent. Four synthetic teen testers ranked them; the main finding was that games where the math and the game have nothing to do with each other lost, and that fun did not equal return visits. It is an experiment feeding the SpeedPrep roadmap, not a product.
-
Client work · Website
BeefShopper
Live
Evidence-first rankings of US home-delivery ribeye, scored by an open algorithm.
Details and links: BeefShopper
BeefShopper compares 270+ ribeye listings across 27 mail-order beef vendors in one searchable table, scored by an algorithm with no editor opinion and sorted by real attributes and price per pound. Every claim carries its source and a verification tier (self-claimed, user-reported, third-party). The site states affiliate commissions never affect scores. Revenue is affiliate links only. It is a narrow, high-intent comparison-site model.
-
Client work · Website
Fable Software
Live
An umbrella brand: software for construction companies where customers own their data.
Details and links: Fable Software
Fable Software is the 'own your data, own your future' front page for a family of field and back-office tools (SitePilot, SolarKnock, SendSignatures and a tracker), aimed at contractors, solar installers and EPC crews. It is positioned against per-seat SaaS pricing and lock-in, and offers a free consultation rather than a signup funnel. It is also the legal entity (Fable Software LLC) behind several products.
-
Client work · Website
SitePilot
Live
Websites and marketing sold as an ongoing, managed growth service rather than a one-time build.
Details and links: SitePilot
SitePilot sells websites and marketing as an ongoing, managed growth service for small and mid-sized businesses: web design, SEO, content, paid ads and AI-driven insight.
-
Client work · Website
TrendZion
Live
A community web app for researching how words and phrases are used among Latter-day Saints.
Details and links: TrendZion
TrendZion lets people search terms used among Latter-day Saints, compare usage trends across up to four themes at once, and comment and vote in a moderated community.
-
Client work · Tool
SendSignatures
Live
A self-hosted, open-source e-signature service positioned as a DocuSign alternative.
Details and links: SendSignatures
SendSignatures is a hosted, Fable Software branded deployment of DocuSeal, the open-source e-signature platform. Send, sign and process documents on infrastructure you control, with an API for developers. Credit for the software goes to DocuSeal.
Nothing matches that search. Try a broader word, or clear the filters.