Work Samples
Dashboards, AI pipelines, and automation systems I've built and run in production. Every card below opens a full case study — the actual architecture, workflow stages, tooling, and safeguards behind it.
These are the same patterns I use when building automations for clients.
Featured Projects
The flagship builds — systems I designed, shipped, and still run in production every day. Each one has a detailed walkthrough.
Financial Portfolio Dashboard
Production SaaS: real-time investment tracking on 230+ serverless handlers
Problem
Monitoring stocks, options, and crypto across multiple brokerages meant logging into five different accounts and still having no consolidated view of real P/L.
What I built
- Multi-portfolio dashboard with live quote polling and after-hours pricing
- 230+ serverless handlers and 31 scheduled cron jobs proxying every data provider
- Options P/L tracking with real-time Greeks for multi-leg spreads
- Brokerage account sync, AI chat assistant, and SEC filing alerts
Stack
Personal educational tool, not financial advice software.
SEC Filing Monitor
Hourly AI filing analysis where Git itself is the database
Problem
Material disclosures land in SEC EDGAR filings at unpredictable times, and nobody has time to poll EDGAR by hand or read a 200-page 10-K to find what changed.
What I built
- Hourly analysis on a small always-on server, with the retired workflow kept as a rollback path
- SEC EDGAR polling with rate-limit and user-agent compliance built in
- AI-scored findings graded for materiality, with a measured model-swap kill switch
- Results committed straight into Git, so the data store has full history
Stack
AI Content Pipeline
Discovery to WordPress draft: a full editorial pipeline running unattended
Problem
Scaling how-to article production across multiple tech blogs was capped by how fast one person could research, write, and screenshot everything by hand.
What I built
- Multi-stage pipeline: discovery → research → outline → draft → screenshots → publish
- Puppeteer automation that navigates real apps to capture accurate UI screenshots
- Topic deduplication against everything already published
- Quality gates before anything reaches WordPress — drafts only, never auto-live
Stack
Gmail Workspace Auto-Response System
Per-site daemons that read inbound customer mail and draft contextual replies
Problem
Inbound customer email across several properties needed fast, on-brand responses — but handing an AI unsupervised access to a customer inbox is a risk no business should accept.
What I built
- Per-site daemons reading inbound mail through the Google Workspace CLI
- A portable AI skill that drafts replies using each site's own tone and context
- Draft-only posture with a human approval gate — nothing sends itself
- State tracking so a thread is never answered twice, plus failure alerting
Stack
AI Tax Preparation Workspace
Reading tax PDFs into reviewable data — with a human kept firmly in the loop
Problem
Preparing personal and business returns means pulling hundreds of individual figures out of PDFs by hand, then working out which forms are still missing — slow, and easy to get wrong in ways you only discover later.
What I built
- Document intake that identifies each form and pulls out its individual values
- Bundled prior-year returns split into component forms by page range
- Every extracted figure held for human review with a confidence score
- Readiness board comparing last year's filing against this year's paperwork
Stack
Internal single-operator tool. Interface visuals are illustrations with placeholder data.
Automation Systems
Unattended systems running on schedules — with the cost controls, approval gates, deduplication, and failure alerting that keep autonomous automation safe to leave running.
Overnight Content Orchestrator
Cost-gated autonomous AI that runs unattended overnight — without runaway bills
Problem
Autonomous overnight AI has two failure modes that terrify clients: a silent break nobody notices for a week, and an API bill that quietly runs into four figures.
What I built
- Tier-0 precheck that runs before any expensive model is ever invoked
- Necessity-gated prompting so it only does work that's genuinely needed
- Guaranteed failure email — silent breakage is structurally impossible
- Runs on a dedicated cron-offload droplet, isolated from production sites
Stack
Serverless Skills Backend
Architecting AI systems so the expensive model does the least possible work
Problem
AI agents are the most expensive part of any pipeline, yet they were being used for deterministic work — fetching data, rendering charts, building email HTML — that cheap code does better.
What I built
- Shared serverless API that absorbed all deterministic work off the AI agent
- Server-side chart rendering, financial data fetching, and image hosting
- Persistent deduplication and run-state so repeat content never ships
- Transactional email delivery with server-built templates
Stack
Social Post Approval Dashboard
Human-in-the-loop review and publishing for AI-generated X and LinkedIn posts
Problem
AI can draft social content faster than anyone can review it — but publishing unreviewed AI output to a public brand account is how companies end up in screenshots.
What I built
- Durable Neon ledger tracking every post through pending → posting → posted
- Atomic claim on publish so concurrent runs can never double-post
- Human edits preserved separately and only promoted once publishing confirms
- Channel-aware publishing for X and LinkedIn, with SHA-256 content dedup
Stack
Multi-Site Monitoring Suite
On-demand health checks for a site portfolio, run from chat instead of a dashboard
Problem
A monitoring dashboard only works if someone logs into it. Checking uptime, SSL, plugin updates, and traffic across a portfolio of sites meant opening a host panel per site.
What I built
- Dashboard consolidating uptime, SSL, plugin/theme, backup, and storage state
- Telegram bot with slash commands for on-demand checks from my phone
- Resilient host API layer with cached auth and partial-outage isolation
- HTTP probes returning structured latency and status instead of throwing
Stack
SaaS Cost & Revenue Dashboard
Unit economics in one view — 13+ vendor costs tracked against real revenue
Problem
Running a SaaS on a dozen metered vendors means the true cost per customer is spread across 13 billing portals and effectively unknowable at a glance.
What I built
- Automatic cost fetching from AI, infrastructure, and data provider APIs
- Per-model AI spend breakdown by input and output tokens
- Revenue analysis joined against cost to surface real margin
- Trend tracking to catch a vendor bill spiking before the invoice lands
Stack
Google Workspace Automation
Automated workflows for Gmail, Sheets, Drive, and Calendar using the Google Workspace CLI.
Automated Report Generation
Weekly business reports built from Sheets, Drive, and Gmail data
Problem
Manual report assembly takes 2+ hours every Monday morning, pulling from multiple Google sources.
What I built
- Multi-source data extraction from Sheets, Drive, and Gmail
- Formatted report generation with metrics and highlights
- Automated email delivery and Drive archival
- Calendar reminder for report review
Stack
Inbox Triage Automation
AI-powered email classification and routing from Gmail to Sheets
Problem
Important emails buried in a cluttered inbox. Invoices, client requests, and follow-ups all mixed together.
What I built
- Email classification by type, sender, and urgency
- Automated routing to categorized Google Sheets
- Calendar reminders for unanswered emails
- Newsletter archival and labeling
Stack
Client Onboarding Pipeline
New spreadsheet row triggers folders, emails, and calendar invites
Problem
Every new client means 15 minutes of manual setup: creating folders, writing emails, scheduling meetings.
What I built
- Trigger-based pipeline from Google Sheets
- Automatic Drive folder structure creation
- Personalized welcome email from templates
- Calendar kickoff scheduling
Stack
Workspace Health Check
Automated audit of stale files, calendar conflicts, and permissions
Problem
Files pile up, calendars get messy, permissions go stale — and nobody notices until something breaks.
What I built
- Drive file activity scanning and stale file detection
- Calendar conflict identification and resolution suggestions
- Unanswered email tracking with priority digest
- Shared drive permissions audit
Stack
Platform & Product Engineering
Longer-lived application and infrastructure work — multi-site architecture, data reconciliation, and browser tooling.
Multi-Site WordPress Theme System
Parent/child architecture where one fix ships to every property
Problem
Running several WordPress sites with shared functionality meant the same bug had to be fixed four separate times, four slightly different ways.
What I built
- Parent theme holding all shared functionality and performance work
- Per-site child themes for branding and layout overrides
- Gulp build pipeline for asset compilation and optimization
- rsync-based deployment workflow across all properties
Stack
Author Performance Dashboard
Laravel app reconciling analytics, CMS, and workflow data into writer payments
Problem
Managing writers across multiple sites meant reconciling post counts, traffic performance, and payment owed across three disconnected systems every month.
What I built
- Full Laravel application with per-author performance reporting
- Google Analytics integration for traffic attribution
- WordPress sync for authoritative post tracking
- Trello integration for editorial workflow state
- Automated payment calculation from reconciled data
Stack
Gmail Inbox Timer Extension
Chrome extension surfacing how long each email has been waiting
Problem
Gmail shows you when an email arrived, but not how long it has been sitting unanswered — so the oldest requests are the easiest ones to forget.
What I built
- Chrome extension injecting wait-time into the Gmail interface
- Resilient DOM parsing against a third-party UI that changes without notice
- Color-coded urgency thresholds for at-a-glance triage
- Settings UI with synced preferences across devices
Stack
AI Build Case Studies
Client-facing AI products and internal tools delivered with clear scope, human review, and production-ready handoff. See also: AI Email & CRM Automation.
Support Inbox Copilot
AI triage assistant for faster ticket routing and first-draft responses
Problem
Support teams were spending too much time sorting tickets and writing repetitive replies, which slowed response times.
What I built
- Intake pipeline to classify tickets by topic, urgency, and intent
- Draft-reply generation using approved tone and policy constraints
- Human approval queue before any response is sent
- Daily metrics dashboard for queue health and response speed
Stack
Client details anonymized.
Sales Call Intelligence Pipeline
From call transcript to CRM updates and action items in minutes
Problem
Sales leadership needed consistent post-call summaries and follow-up tasks, but manual note review was too slow.
What I built
- Automated transcript ingestion from recorded calls
- AI-generated summaries, objections, and next-step recommendations
- Structured updates pushed into CRM fields and tasks
- Slack digest with high-priority deals and risk flags
Stack
Client details anonymized.
Proposal Draft Generator
Internal app that turns discovery notes into first-pass proposals
Problem
Teams were rewriting proposal language from scratch for each new lead, creating delays and inconsistent quality.
What I built
- Structured intake form for project goals, scope, and constraints
- AI draft generation using reusable proposal templates
- Editable review screen with version history
- Export flow for client-ready docs and internal approvals
Stack
Client details anonymized.
More examples available on request
Some projects are under NDA or involve sensitive business data. If you're evaluating whether I can handle a specific type of project, let's talk.
Want to discuss a similar project?
If one of these looks close to what you need, the fastest path is a short call about your specific workflow.
Get in Touch