Selected Work

What we've built.

Real AI systems, shipped and running. Here is what they replaced, what they do now, and how they were made.

Systems shipped

AI agents, one cockpit

Tests, revenue engine

Book a Call

Multi-Tenant STR SaaS

STR Secrets

“Five AI agents per operator, one multi-tenant dashboard, running every night.”

The Problem

Operators were stitching together spreadsheets, PMS exports, and guesswork. STR Secrets puts a full AI operations team behind one login, isolated and secure per workspace.

The Build

A multi-tenant SaaS where each short-term-rental operator connects their PMS and tools, then five specialist AI agents (Revenue, Guest, Operations, Analytics, Marketing) analyze pricing, messaging, turnovers, and market data, coordinated by an orchestrator, streamed live, and swept nightly.

Before

Pricing, messaging, and ops run by hand
Data scattered across PMS and spreadsheets
No isolation or security for client data
Insights that are always a day behind

After

Five specialist agents run the operation
One dashboard connects PMS, pricing, and market data
Per-tenant encrypted credentials, RLS isolation
Nightly automation plus live run streaming

How it was built

Built a strict multi-tenant backend

Express and Supabase Postgres with row-level security, a 5-layer architecture, and AES-256-GCM per-tenant credentials.

Coordinated five agents with an orchestrator

Revenue, Guest, Operations, Analytics, and Marketing agents run on schedule and stream every run live over Socket.io.

Shipped as a real product

React 19 dashboard, Stripe billing, and CI/CD that auto-deploys on merge to main.

React 19Node 22ExpressSupabaseSocket.ioStripeOpenRoutern8n

AI agents per operator

Nightly

automated sweep + live stream

Isolated

per-tenant, RLS-enforced

Live

in production, CI/CD

AI Operations Platform

LegacyRnR Control Center

“Nineteen AI agents for one property-management operation, under a single cockpit.”

The Problem

A growing property-management operation had 19 distinct jobs to automate and no safe way to run AI against live guest and reservation data. The cockpit runs every agent with real logic and human approval gates.

The Build

A unified command center for LegacyRnR that runs 19 specialist AI agents on top of Guesty: guest, owner, and operations agents grouped into orbits, with a dual-approval gate on anything touching money or compliance, an immutable activity log, and three external portals.

Before

19 manual jobs across a property operation
No safe way to run AI on live guest data
No audit trail for automated actions
Owners and guests on disconnected tools

After

All 19 agents run under one command center
Guesty wired live for real listings and reservations
Dual-approval queue: nothing auto-applies on money
Immutable activity log and three external portals

How it was built

Turned specs into 19 real agents

Built from the client's agent build list into a typed agent runtime with training briefs and real run logic, no mock data.

Wired Guesty live behind a snapshot adapter

A cached snapshot maps live listings and reservations into every agent, with writes gated behind approval.

Gated every sensitive action

A dual-approval queue and append-only activity log so money and compliance actions are human-approved, never auto-applied.

Next.js 14TypeScriptSupabaseInngestClaudeGuesty

AI agents, one cockpit

Guesty

wired in live

Dual

approval on money + compliance

guest, owner, education portals

STR Management Automation

West Properties Ops Platform

“Eight AI agents running the busywork of a 27-property management company.”

The Problem

A management company with revenue-share partners was losing its team to reporting, messaging, and copy-paste work between tools. The platform runs those jobs automatically, with a human approval loop in Slack before anything reaches a partner or guest.

The Build

A single-client operations platform for West Properties: eight specialist agents covering guest messaging, review intelligence, partner statements, revenue sheets, an AI revenue manager on top of PriceLabs, daily booking reports, a Slack-to-Monday bridge, and a content engine, wired into Guesty for Pros, per-partner QuickBooks, and Google Sheets.

Before

Partner statements assembled by hand every month
Guest messages and reviews handled one by one
Pricing managed ad hoc across 27 properties
Team copy-pasting between Slack, Monday, and Sheets

After

Statements generated and routed for approval automatically
Guest messaging and review intelligence run as agents
Daily and weekly AI pricing scans on top of PriceLabs
Slack and Monday bridged, reports delivered daily

How it was built

Built a strict layered backend

Adapters, controllers, services, and repositories with per-partner QuickBooks OAuth and Supabase Vault for secrets.

Shipped agents behind approval loops

Each agent runs in shadow mode first; statements and pricing pushes require named-human approval in Slack before anything sends.

Automated the calendar

n8n crons drive daily 30-day and weekly 90-day pricing scans, booking reports, and statement cycles.

NodeExpressSupabaseGuestyPriceLabsQuickBooksSlackn8n

AI agents, one platform

properties under management

tests behind the platform

Slack

approval loop before anything sends

STR Operations Platform

Stay With Somos Platform

“Seven AI agents and one dashboard for a mixed arbitrage-and-management portfolio.”

The Problem

Operators running both arbitrage units and managed properties get statements wrong with off-the-shelf tools: flat rent lines for one model, commission and expenses for the other. This platform models both per property, per owner.

The Build

A full operations platform for Stay With Somos: seven specialist AI agents (guest messaging, revenue intelligence, analytics, operations, marketing, market comping, and lead response) plus an accounting engine posting across two QuickBooks companies, all behind a seven-page dashboard, with owner statements that correctly handle a portfolio mixing rental arbitrage and managed properties.

Before

Two QuickBooks companies reconciled by hand
Owner statements broke on mixed business models
Guest messaging and pricing run manually
Leads followed up whenever someone had time

After

Accounting agent posts across both ledgers
Per-property statements: flat rent or commission, correctly
Seven agents run the day-to-day operation
Lead agent responds and follows up automatically

How it was built

Integrated the full stack

Guesty, dual-realm QuickBooks, PriceLabs, GoHighLevel, Stripe, and Turno wired in behind one backend.

Modeled the mixed portfolio

Per-property business-model rules so arbitrage units bill flat monthly rent while managed units carry commission and expenses.

Ran shadow mode before cutover

Every agent message and posting visible for review before the system went live with real guests and real money.

NodePythonSupabaseGuestyQuickBooksPriceLabsGoHighLevelStripe

AI agents behind one dashboard

QuickBooks companies, one view

Mixed

arbitrage + managed statements

dashboard pages, live data

STR Revenue Automation

Solnest Automation Stack

“Revenue intelligence, pricing reports, and market scraping, on autopilot.”

The Problem

Revenue management and market research for short-term rentals is hours of manual scraping and spreadsheet work every week. This stack runs it automatically and delivers decisions, not raw data.

The Build

The automation stack powering Solnest Stays: a multi-agent revenue engine, a PriceLabs agent that scrapes rates weekly and emails AI pricing reports, plus Airbnb and Instagram scrapers that feed Claude analysis, all wired through n8n.

Before

Hours of manual PriceLabs and market research weekly
Pricing decisions made on gut feel
Competitor and review data pulled by hand
No repeatable revenue process

After

A battle-tested revenue engine makes the calls
PriceLabs scraped weekly, AI reports emailed automatically
Airbnb and Instagram data pulled via Apify, analyzed by Claude
One repeatable, automated revenue workflow

How it was built

Built a tested revenue engine

A Python pricing engine of 12 modules backed by 164 tests turns market data into pricing decisions.

Automated the weekly pricing report

A Node agent scrapes PriceLabs, generates an AI pricing report, and emails it, no manual pull required.

Fed the agents live market data

Apify scrapers pull Airbnb listings, reviews, and Instagram profiles into Claude for analysis.

ClaudePythonNode.jsPriceLabsApifyn8n

tests behind the revenue engine

pricing decision modules

Weekly

AI pricing reports, emailed

Claude

analysis on scraped market data

MedSpa Automation

Patient Concierge Agent

“87% booking rate from inbound leads. Fully hands-free.”

The Problem

The front desk was losing 40% of leads to voicemail during peak hours. The concierge agent now captures every inquiry within seconds and books them directly into the calendar.

The Build

An AI concierge that handles appointment scheduling, treatment consultations, pre-care instructions, and post-treatment follow-ups for a luxury MedSpa - across SMS, email, and web chat simultaneously.

Before

40% of leads going to voicemail during peak hours
2-person front desk overwhelmed by phone + walk-ins
Follow-ups falling through the cracks
$18K+ in potential revenue lost monthly

After

Every inquiry captured in under 3 seconds
87% of leads converted to booked appointments
Automated pre-care and post-treatment follow-ups
Front desk focuses on in-person experience only

How it was built

Audited the lead funnel

Tracked where inquiries came from and where they dropped off - voicemail was the #1 killer.

Built multi-channel concierge

SMS, email, and web chat all handled by one AI agent with full treatment knowledge.

Connected to Cal.com

Appointments booked directly into the clinic's calendar - no human handoff needed.

Claude AITwilioCal.comn8nAirtable

Lead-to-booking rate

2.4x

More bookings/mo

Missed inquiries

$0K

Monthly revenue added

MedSpa Finance AI

MedSpa Financial Automation

“The monthly financial close of a med spa, fully automated with an AI CFO.”

The Problem

Clinic owners were losing 3-4 days a month to manual bookkeeping across Jane, RepeatMD, and Stripe - and still had no clear picture of the numbers. The close now runs itself, with a human approving before anything posts.

The Build

A financial automation system that replaces days of fractional-CFO work every month: bots collect data from the booking, membership, and payment systems, journal entries post to QuickBooks automatically, an 18-KPI engine feeds a nine-page dashboard, and Claude writes plain-English CFO commentary delivered as a monthly report. Now running for two clinics.

Before

3-4 days of manual close work every month
Data scattered across Jane, RepeatMD, and Stripe
Hundreds of SKUs reconciled by hand
Numbers arrived too late to act on

After

Automated collection, matching, and posting
18 KPIs and a 9-page dashboard, always current
AI CFO commentary explains the month in plain English
Human-in-the-loop approval before anything posts

How it was built

Automated the data collection

A Playwright bot pulls Jane reports; RepeatMD and Stripe flow in through APIs, all landing in Supabase via n8n.

Built the accounting agents

Expense review, GL validation, deposit matching, and vendor-bill agents - each with confidence scoring and human approval before QuickBooks posting.

Shipped the AI CFO layer

An 18-KPI engine feeds a React dashboard and a Claude-written monthly commentary, delivered as a PDF report.

n8nSupabaseQuickBooksStripePlaywrightClaudeReactGoHighLevel

KPIs computed monthly

journal entries auto-posted to QuickBooks

dashboard pages, live financials

clinics running in production

Restaurant AI

Voice Ordering Agent

“Handles 120+ calls per day. Never puts anyone on hold.”

The Problem

The restaurant was losing $3,200/week in abandoned phone orders during the dinner rush. Staff couldn't answer fast enough. The voice agent now handles the entire queue.

The Build

A voice AI agent that takes phone orders for a high-volume restaurant - understanding menu customizations, dietary restrictions, upselling combos, and processing payments. Speaks naturally with under 400ms latency.

Before

$3,200/week lost to abandoned phone orders
Customers on hold 4+ minutes during dinner rush
Staff pulled from kitchen to answer phones
Order errors from rushed, distracted staff

After

120+ calls handled daily with zero hold time
94% order accuracy - better than human staff
$3,200/week in revenue recovered immediately
Kitchen staff stays in the kitchen

How it was built

Mapped the full menu + edge cases

Every item, modifier, combo, allergy note, and upsell - programmed into the voice agent.

Built natural voice flow via Vapi

Under 400ms response latency. Customers don't realize they're ordering from AI.

Integrated with Square POS

Orders go straight into the kitchen queue. Payment processed on the call.

VapiClaude AISquare POSn8nTwilio

Daily calls handled

Order accuracy

$3.2K

Weekly revenue recovered

0.4s

Voice latency

Dental Clinic AI

Patient Voice Agent

“Zero missed patient calls. 24/7 scheduling that sounds human.”

The Problem

The clinic's two-person front desk was overwhelmed, resulting in 35% of calls going to voicemail. Most of those patients never called back. The voice agent eliminated that entirely.

The Build

A conversational voice AI that answers every call to a dental clinic - scheduling appointments, handling insurance pre-qualification questions, sending appointment reminders, and managing cancellations and rebookings automatically.

Before

35% of patient calls going to voicemail
Most voicemail patients never called back
High no-show rate with no automated reminders
28 hours/week spent on phone scheduling alone

After

0% missed calls - every patient gets answered
62% reduction in no-shows via smart reminders
28 hours/week freed up for patient care
4.9★ patient satisfaction rating maintained

How it was built

Analyzed call patterns

Mapped peak call times, common questions, and where the front desk bottlenecked.

Built voice agent with Vapi

Natural conversation flow for scheduling, insurance questions, and reminders.

Connected to Dentrix

Direct integration with the clinic's practice management software - real-time availability.

VapiClaude AIDentrix APIn8nTwilio

Missed calls

Fewer no-shows

Hours saved/week

4.9★

Patient satisfaction

Real Estate AI

Lead Gen Agent

“Every lead answered in under 60 seconds. Around the clock.”

The Problem

The average agent takes over 15 hours to respond to a new lead, and 78% of buyers end up working with whoever answers first. Responding inside 5 minutes makes a lead 21x more likely to qualify - so the agent responds in seconds, every time, including nights and weekends.

The Build

An AI lead-response agent for a real estate team that engages every inbound lead instantly over SMS and email - qualifying budget, timeline, and area, scoring the lead, booking showings into the agent's calendar, and running the full follow-up sequence until the lead answers or opts out.

Before

First response measured in hours, not seconds
Leads arriving nights and weekends went cold
Follow-up stopped after one or two attempts
No qualification - every lead treated the same

After

Every lead engaged in under a minute, day or night
Budget, timeline, and area qualified automatically
Scored leads pushed to the CRM, showings booked
Persistent follow-up until answer or opt-out

How it was built

Wired every lead source into one intake

Zillow, portal, and website leads all flow into a single pipeline - nothing depends on an agent seeing an email.

Built instant qualification over SMS

The agent opens a natural conversation, qualifies budget, timeline, and area, and scores the lead in real time.

Automated booking and follow-up

Qualified leads book straight into the calendar via GoHighLevel; everyone else enters a persistent multi-touch sequence.

Claude AIGoHighLevelTwilion8n

<0s

Response time, 24/7

Higher qualification odds vs. 30-min response

Follow-up touches, automated

Leads left cold

What We Build

Six kinds of systems, one team.

AI agents & orchestration

Specialist agents coordinated by an orchestrator, each with its own job and its own scope.

Proven in — STR Secrets · LegacyRnR Control Center

Revenue & pricing intelligence

Market data turned into pricing decisions, tested and repeatable, not guesswork.

Proven in — Solnest Automation Stack

Guest & customer messaging

Conversations handled automatically, on brand, without dropping a single thread.

Proven in — Patient Concierge Agent

Multi-tenant SaaS platforms

Isolated, secure workspaces built to hold real customer data at scale.

Proven in — STR Secrets

Automation & data pipelines

Scraping, syncing, and scheduling that runs on its own clock, not yours.

Proven in — Solnest Automation Stack

Approval-gated operations

Anything touching money or compliance waits for a human, logged and reversible.

Proven in — LegacyRnR Control Center

Real Tools, Real Builds

ClaudeOpenAIn8nSupabaseStripeGuestyPriceLabsApifySocket.ioPostgresVercelGitHubClaudeOpenAIn8nSupabaseStripeGuestyPriceLabsApifySocket.ioPostgresVercelGitHub

Want something like this for your business?

Book a Call

See services →