Projects
Systems we've built
Client work stays confidential. These reference systems show how we build: tested, documented, and with the security controls built in from day one.
Reference build
AI Phone Agent with Fraud Controls
A voice agent that answers a real phone line, books meetings, texts payment links, and opens tickets.
The problem
Most AI phone agents can take actions but have no limits. A confused agent, or a caller trying a scam, can book ten meetings or send repeat charges.
What it does
- Answers calls through Twilio using OpenAI Realtime, Claude, or Gemini
- Books meetings, sends secure Stripe payment links by text, opens Zendesk tickets, saves leads to a CRM
- Writes a short summary of every call into the CRM when the caller hangs up
Reference build
Private AI for Sensitive Documents
A private AI model on your own hardware that answers questions about your files, with citations.
The problem
Law, medical, and finance offices want AI but can't send client files to a public AI service.
What it does
- Runs open models locally with installers for Mac, Windows, and Linux
- Upload PDFs and documents, then ask questions and get answers that cite their sources
- Works with any app that already supports the OpenAI API
Reference build
AI Cost Control Gateway
One connection to OpenAI, Anthropic, Gemini, and local models, with spending caps and alerts.
The problem
AI bills grow quietly. A bug or a leaked key can burn through a month's budget in an hour.
What it does
- Routes each request to the best model and falls back if a provider is down
- Caches repeated answers and reports the dollars saved
- Live dashboard of spend by app and team
Want one of these for your business?
Each of these can be adapted to your tools, your data, and your rules. Start with a free call.