Systems in production, with the numbers behind them.

The problem

Hundreds of tickets a day arrived across guest messages, SMS, email and internal apps. First response took more than 24 hours, and most of the process lived in people's heads rather than in writing.

What we built

• An agent that works each ticket end to end in chat and voice: answers, status updates, check-in help and follow-ups
• Answers grounded in the client's own SOPs and knowledge base
• Work orders created automatically, with category, priority and unit already filled in
• An escalation engine that weighs confidence, off-topic requests, edge cases and sentiment before handing off
• Hourly health checks that raise an alert the moment anything stops

The result

The team no longer triages the inbox by hand. Routine tickets close without a person, and anything uncertain reaches a human with the full context attached, because an escalation is always better than a bad answer.

The problem

Back-office teams spent a large part of every week checking, matching and re-keying documents by hand, and they wanted AI built into their daily work rather than one more tool to log into.

What we built

• Document-processing pipelines that read and check incoming paperwork
• Reconciliation tools that match records across systems automatically
• AI assistants connected to internal systems and data through custom connectors
• Customer-facing AI assistants built into the company's products
• Hands-on rollout with each team so the tools became part of the routine

The result

Capacity went up about 35% in under two months. The document and reconciliation systems, delivered in about six weeks, removed $50k+ a year of manual work on their own, and they are only one part of what was built.

The problem

The sales team spent most of its week prospecting by hand, stitching property data, a contact database and the CRM together before a single message went out.

What we built

• Property data pulled from CoStar, then segmented and scored by fit
• Contact enrichment through ZoomInfo
• A custom-object HubSpot setup, loaded automatically
• Owner and manager lists for the top 20 metro areas, ready for outreach

The result

Sales starts the week with scored, enriched contacts already in HubSpot and spends its time on conversations instead of research. The lists have already opened six figures of new sales pipeline.

The problem

Most homeowners only service their equipment once something breaks, and then they are left searching for a contractor and hoping the one they find is reliable.

What we built

• An SMS concierge that remembers every system, appliance, install date and service record in the home
• Maintenance reminders timed to the actual equipment
• Troubleshooting that tells a quick DIY fix apart from a job for a licensed pro
• Vendor scheduling, dispatch and payments
• A web app, plus a dashboard for partner businesses

The result

The platform went from an idea to production with live users, and it runs entirely over text, so homeowners never have to install an app.

The problem

Clinics run on many niche practice-management systems, each with its own API, so every new clinic meant another custom integration before an agent could book anything.

What we built

• A master MCP server in n8n that routes each request to the right system adapter
• Per-clinic settings stored in Supabase, so adding a clinic is configuration, not new code
• Voice agents on ElevenLabs and chat agents in n8n that book through it
• A patient billing form and a payment agent connected to the clinic CRM

The result

Agents book through one shared layer instead of one integration per clinic, and patients get answers by voice or chat at any hour.