Projects

From AI orchestration to production apps—building solutions that solve real problems

★ Featured

Gastown

AI agent orchestration framework for parallel multi-project development

Custom FrameworkCLINode.jsJSON ConfigurationAI Agent Orchestration

Problem

Traditional AI-assisted development suffers from sequential bottlenecks: one AI conversation at a time, manual context-switching between projects, repeated explanations of project structure, and context pollution from long conversation histories. Developers managing multiple projects waste valuable time juggling contexts and waiting for one task to finish before starting another. The single-threaded nature of AI assistants becomes a productivity ceiling.

Solution

Gastown is an innovative AI agent orchestration framework that enables true parallelization across multiple projects. Named after frontier towns where specialized workers operate autonomously, Gastown coordinates distributed AI "workers" (called polecats) across independent project workspaces (called rigs). It treats AI agents as ephemeral, single-task contractors that spawn fresh with exactly the context they need, execute work autonomously, and disappear—enabling simultaneous development across an entire project portfolio.

Multi-Project Orchestration: Coordinate work across multiple independent projects (rigs) simultaneously—wheeltracker, mealplanner, and personalsite are current active rigs
Ephemeral Worker Model: Polecats are disposable AI agents that spawn fresh for each task with zero context pollution, execute work, and self-destruct
The Propulsion Principle (GUPP): Revolutionary autonomous execution—if work appears on an agent's "hook," they execute immediately without waiting for approval, treating the system like a steam engine
CLAUDE.md Context Injection: Each rig maintains a master context file automatically injected into every polecat, ensuring consistent quality without repetition
Convoy Pattern: Batch related tasks for parallel execution with unified progress tracking across features
Mayor Coordination: Central orchestration agent breaks down PRDs into distributable work units and dispatches to polecats
Automated Integration Pipeline: Refinery agent processes merge queue with code review and verification before integration
Distributed Monitoring: Deacon (town-level health), Witness (per-rig monitoring), and Daemon (background services) ensure system reliability
Structured Work Management: Beads system for granular task tracking with cross-rig coordination

Architecture & Design

Architecture: Distributed workforce model with ephemeral workers, persistent context, and event-driven dispatch
Agent Roles: Mayor (coordinator), Polecats (workers), Deacon (health monitor), Witness (activity tracker), Refinery (code reviewer), Daemon (background service)
Work Distribution: PRD → Mayor → Convoys → Polecats → Refinery → Integration
Session Management: Mandatory completion protocol (tests, commit, push, quality gates)
CLI Interface: gt command suite for rig management, polecat spawning, mail/messaging, work dispatch
Configuration: JSON-based (town.json, rigs.json, daemon.json) with per-rig CLAUDE.md context files

Frontend

Next.js 15+
React 19
TypeScript
Tailwind CSS

Backend

Node.js
PostgreSQL
Prisma ORM

Testing

Vitest
Playwright

Deployment

Multiple production rigs: wheeltracker, mealplanner, personalsite
Mandatory completion protocol with tests, commit, push, quality gates

Multiple AI agents working simultaneously across different projectsTrue Parallelization

Zero context pollution—each polecat starts with exactly what it needsFresh Context

GUPP principle enables continuous progress even when humans are AFKAutonomous Execution

Read Blog Post View Source

Wheel Tracker

Sophisticated options trading tracker for the wheel strategy

Next.jsReactTypeScriptPostgreSQLPrismaTailwind CSS

Problem

Options traders using the wheel strategy need a comprehensive tool to track trades across multiple cycles, calculate complex profit/loss metrics, manage assigned positions, and benchmark performance against market indices. Existing solutions are either too complex, too simplistic, or don't understand the nuances of the wheel strategy.

Solution

Wheel Tracker is a production-ready Next.js application that provides end-to-end tracking for options traders. It manages the complete wheel cycle—from selling cash-secured PUTs, to managing assigned stock positions, to selling covered CALLs—with sophisticated P&L calculations, real-time price tracking, and performance analytics.

Complete Wheel Cycle Tracking: Manages PUTs, stock positions, and CALLs with relationship tracking between trades
Advanced Analytics: Real-time P&L calculations (realized vs. unrealized), win rate tracking, and performance metrics by ticker
Expiration Calendar: Color-coded ITM/OTM/ATM status with batch operations for managing upcoming expirations
Market Benchmarking: Compare portfolio performance against SPY, QQQ, and VTI indices
Live Price Integration: Real-time stock prices via Alpha Vantage API with automated daily updates
Tax-Ready Exports: Export trade history to CSV for tax preparation

WCAG accessibility complianceProduction Ready

Zero Cumulative Layout Shift for optimal UXZero CLS

409+ passing testsTest Coverage

View Live Demo View Source

Meal Planner Agent

AI-powered meal planning that saves time and reduces decision fatigue

Next.jsReactTypeScriptClaude AIPostgreSQLRedisBullMQ

Problem

Busy professionals, families, and health-conscious individuals face the daily challenge of "what's for dinner?" Traditional meal planning is time-consuming, requires nutritional knowledge, and becomes repetitive. People need a solution that automates meal planning while respecting dietary preferences, nutrition goals, and household dynamics.

Solution

Meal Planner Agent is an AI-powered web application that leverages Claude AI to generate personalized, nutritious weekly meal plans. It automates the entire meal planning process—from recipe generation to shopping list creation—while adapting to individual preferences, dietary restrictions, and nutritional targets.

AI-Powered Generation: Uses Claude Sonnet 4 to create personalized meal plans with detailed recipes and nutrition information
Household-Based Planning: Share meal plans across family members while maintaining individual nutrition preferences
Smart Shopping Lists: Automatic ingredient aggregation, categorization, and HEB grocery integration with product links
Recipe Favorites System: Bookmark and save favorite recipes for quick reuse
Automated Email Delivery: Beautifully formatted meal plans sent to all household members automatically
Scheduled Generation: Configure automatic weekly meal plan generation on preferred days
Nutrition Optimization: Set protein targets, calorie limits, and dietary restrictions (vegetarian, vegan, gluten-free)
Real-Time Progress: Live progress tracking (0-100%) during meal plan generation

Architecture & Design

Anthropic SDK with Claude Sonnet 4
Structured JSON output
Context-aware generation
Puppeteer for HEB web scraping
Monorepo structure (web app, core agent logic, database, queue worker)

Frontend

Next.js 14
React 18
TypeScript
Tailwind CSS

Backend

Node.js
PostgreSQL
Prisma ORM
Redis
BullMQ job queue
Nodemailer with Gmail SMTP

Testing

E2E testing
Unit testing
Integration testing

Deployment

Multi-tenant architecture with data isolation
BullMQ with rate limiting (10 jobs/min)
Concurrent processing
Persistent storage
Google Analytics 4

Complete data isolationMulti-Tenant

E2E, unit, and integration testingProduction Grade

Job failure recovery and queue managementScalable Workers

View Live Demo View Source

Precept Tracker

Daily Zen journal for reflecting on the 16 Precepts of Soto Zen Buddhism

ElectronReactTypeScriptSQLiteFramer MotionVite

Problem

Practicing the 16 Precepts of Soto Zen Buddhism requires consistent daily reflection, but there are no dedicated tools to support this discipline. Practitioners rely on generic journaling apps or pen-and-paper methods that don't understand the structure of the precepts—the Three Refuges, Three Pure Precepts, and Ten Grave Precepts—making it difficult to track progress, maintain streaks, and observe patterns in their practice over time.

Solution

Precept Tracker is a desktop application purpose-built for daily Zen practice. It guides users through structured reflection on all 16 precepts with customizable prompts, tracks practice streaks and meditation sessions, and provides weekly analytics to reveal trends in your practice. All data stays local on your machine via SQLite.

Guided Daily Reflection: Structured prompts for all 16 precepts organized into Three Refuges, Three Pure Precepts, and Ten Grave Precepts
Practice Streak Tracking: Monitor daily practice consistency with streak counters and history
Meditation Tracking: Separate tracking for meditation sessions alongside precept reflection
Weekly Analytics: Charts and trend visualization to observe patterns in your practice over time
Customizable Prompts: Personalize reflection prompts for each precept to match your practice
Data Export: Export your journal data for backup or external analysis
Weekly Nudges: Configurable reminders to support consistent practice

Cross-platform via Electron (macOS, Windows, Linux)Desktop App

All data stored locally with SQLite—your practice stays privateLocal-First

Complete coverage of Soto Zen Buddhist precepts16 Precepts

View Source

Projects

Gastown

Problem

Solution

Key Features

Technical Highlights

Architecture & Design

Frontend

Backend

Testing

Deployment

Wheel Tracker

Problem

Solution

Key Features

Technical Highlights

Architecture & Design

Frontend

Backend

Testing

Deployment

Meal Planner Agent

Problem

Solution

Key Features

Technical Highlights

Architecture & Design

Frontend

Backend

Testing

Deployment

Precept Tracker

Problem

Solution

Key Features

Technical Highlights

Architecture & Design

Frontend

Backend

Deployment