Practical AI systems
I help service businesses move beyond generic chatbots and disconnected AI tools. The work starts with the real process, then adds searchable knowledge, assisted communication, routing, or pilot automation only where it creates a measurable advantage.
The problem I solve
AI adoption stalls for operational reasons, not technical ones. Here's what I hear — and what I build to fix it.
| What your team says | What I build |
|---|---|
| "We adopted AI tools but nobody uses them consistently" | Documented SOPs and workflow maps that show exactly where AI fits into each process — so it's a habit, not a one-off experiment. |
| "Our AI outputs feel generic and don't sound like us" | Prompt engineering and persona-driven AI assistants (Gemini Gems, Claude Projects) trained on your own documentation — so the output reflects your voice and constraints. |
| "Critical knowledge lives in one person's head and AI can't access it" | NotebookLM knowledge bases built from your existing Google Docs, SOPs, and process files — searchable by anyone, cited from primary sources. |
| "We have five AI tools and none of them connect to our actual workflows" | Make automation scenarios that wire AI tools into your existing stack — so outputs route to the right place, trigger the right next step, and don't create manual cleanup work. |
| "We want to use AI, but we do not want another expensive experiment" | A narrow pilot with defined inputs, human checkpoints, success metrics, and an honest go/no-go decision before production investment. |
Relevant proof
Assisted communication, grounded knowledge retrieval, and a safety-first voice-agent pilot. Each case study includes the underlying workflow and full PDF.
AI-Assisted Communications · Project 01
Built a Make-orchestrated onboarding pipeline where Gemini drafts personalized customer emails using the customer's own variables — property type, use case, install window — so automated comms read like a human wrote them. The workflow routes a Tally form submission through Sheets, AI-drafted Gmail, and Calendar in ~2 seconds.
Business value: faster acknowledgement without giving up context or human review.
Knowledge Enablement · Project 02
Converted 85 years of undocumented tribal knowledge into five structured SOPs, a NotebookLM knowledge base, and a persona-driven Gemini Gem assistant ("Dale") that answers operational questions only from the cited source documents — not from general AI knowledge.
Business value: institutional knowledge becomes queryable, cited, and available without interrupting the same expert repeatedly.
AI Pilot Design · Project 03
Scoped a 24/7 AI-assisted voice agent pilot with four explicit safety guardrails, six measurable KPIs, and a go/no-go decision framework. The spec leads with the conditions under which the AI should not respond.
Business value: test capacity and after-hours capture without hiding safety, escalation, or operational risk.
Tooling
The stack stays intentionally practical: tools the business can understand, source material it owns, and a fallback when AI should not decide.
Good first project
Good candidates include an internal knowledge assistant, assisted customer-response workflow, inbox triage system, or a pilot specification for a higher-risk agent.