AI-Assisted Product, Systems, and Full-Stack Builder

Building AI-assisted tools that feel intentional instead of theoretical.

Building internal tools, assistant systems, and Linux workflows with AI as a practical multiplier.

I care most about tools that are actually usable: fast enough, clear enough, and maintainable enough to survive beyond the first demo.

Explore projects Read the story behind the work

Now / August 2026Building Agent Pipeline, a local-first multi-agent coding pipeline running entirely on my own hardware. It is not finished, and the debrief explaining why it kept failing is the most useful file in the repo.

Featured work

Five case studies that show how I use AI differently depending on the problem.

One project uses AI directly in the product. Another uses AI to manage a complex internal software roadmap. Another turns Linux customization into reproducible tooling. Another turns dockerized local models into a voice assistant with streaming speech and native runtime clients. And one is still unfinished: a local-first agent pipeline I am building now, shown with its failures intact.

Work in progress

Agent Pipeline

Featured

A local-first multi-agent coding pipeline I am still building: six model stages that plan, test, implement, review, and judge a task, running by default entirely against local Ollama models on my own hardware. It does not reliably finish real tasks yet, and the most useful file in the repo is the debrief explaining why.

Six pipeline stages run fully local by default against ollama:gpt-oss:20b; frontier CLIs are opt-in, not the default path
A backlog loop driver drains tasks one at a time and installs as a Linux cron job for unattended runs
Pre-flight plan linting blocks unconvergeable plans before any code is written instead of looping to exhaustion
Still work in progress: mechanically robust, not yet reliable end to end on local models

PythonOllamaLinux cron

Read case study

Shipped project

NewForm AI

Featured

A scheduled reporting app that turns ad-platform performance data into readable, AI-generated insight reports with web, email, and PDF delivery paths.

Prompting is limited to the metrics the user selected
HTML, email, and PDF outputs share the same reporting pipeline
Recurring schedules and manual reruns are both supported

Next.js 16TypeScriptTailwind CSS

Read case study

Active internal product

TripInvoiceTracker

Featured

An internal transportation operations app that now runs the whole loop: trip entry, invoicing, payments, PO batching, an audit trail, a public client request portal, and a serverless deployment path.

Public client trip-request portal and PO intake feed a staff approval inbox before anything becomes a real trip
Grouped PO batches keep separate reservations tied together under one shared invoice
Reservation audit trail and activity log make operational changes reviewable after the fact

Node.jsExpressMongoDB

Read case study

Active systems toolkit

CachyOS Config

Featured

A Linux desktop configuration repo that evolved into a reproducible restore kit with custom Waybar modules, popup tooling, and scoped backup management.

Custom control center, system monitor, clock, calendar, and AI-usage Waybar modules
bootstrap.sh restores curated packages and config targets
Backup manager GUI supports whole-app and component-level backup selection

PythonBashWaybar

Read case study

Active AI voice assistant

Jarvan

Featured

A private AI voice assistant built on local models, now with a containerized backend, two documented Proxmox deployment paths, native shell clients on three desktops, and a SwiftUI iOS push-to-talk app.

Docker Compose brings up Ollama, Redis, SearXNG, and the FastAPI assistant backend as one local stack
Native SwiftUI iOS push-to-talk client alongside the Windows, macOS, and Linux shells
Whisper, Piper, wake-word detection, and streaming turn handling work together as one speech pipeline

PythonFastAPIOllama

Read case study

How I use AI

AI as product capability

When AI belongs in the product, I wire it into a real workflow. NewForm AI is the clearest example: selected metrics become a constrained OpenAI request that returns a readable summary for reports and email delivery.

How I use AI

AI as execution leverage

I use AI to think through architecture, reduce uncertainty, and keep complex work documented. In operational software like TripInvoiceTracker, that means using it to accelerate planning, review, and iterative delivery without pretending roadmap work is already shipped.

How I use AI

AI as a system to interrogate

Jarvan and Agent Pipeline are where I stress-test local models in real workflows. Jarvan became a lab for containerized LLMs, wake-word handling, and latency tradeoffs. Agent Pipeline is the harder question: what happens when you chain six local models together and one of them is the weakest link. The answer so far is written down in a debrief rather than claimed as a win.

How I work with AI

The bootstrapper behind how I work with AI

I do not just use AI out of the box. I built a portable repo scaffold that gives Codex, Claude, and Cursor the same local rules, planning structure, and implementation breadcrumbs so the tools fit the way I actually work.

The goal is to manage ideas and ambitious projects with better continuity. Instead of losing context across chats or tools, I keep an append-only record of plans, scope changes, verification, and lessons learned so future AI sessions can improve from prior work instead of starting cold.

AGENTS.md.agent-governance-manifest.jsonscripts/sync-agent-rules.shscripts/create-implementation-doc.shplans/README.mdBOOTSTRAPPER_SPEC.md