Turning technology strategy into business outcomes.

Lori's Voice Management Hub

Lori's Voice needed a dedicated way to manage donors, campaigns, and receipting instead of stretching a Salesforce instance that isn't scoped for it. I'm building a purpose-built donor management platform on Next.js 16, React 19, and TypeScript, backed by Supabase for Postgres, auth (with MFA), storage, and edge functions.

It tracks one-time and recurring donations, campaign attribution, workplace-matching and memorial gifts, and reconciles donations into bank deposits. Supabase edge functions auto-generate IRS-compliant tax receipts and year-end giving statements, and built-in reporting covers donation trends, lapsed donors, and CSV/Excel exports.

Google Gemini assists volunteers in drafting donor emails, thank-you letters, and social posts from prompt templates, with Resend handling transactional email and a Playwright/Vitest suite covering the app end to end. The goal: a lightweight, tailored alternative to a general-purpose CRM, right-sized for a volunteer-run nonprofit's actual fundraising workflow.

Agentic Workflows

I'm hands-on with frontier agentic tools — Claude Code, Google Antigravity, and ChatGPT — plus open-source harnesses like Hermes and OpenClaw, which can tap AI compute from either the cloud or self-hosted local models. Rather than one-off prompts, I'm learning how these tools chain multiple steps together to create workflows, and evaluating where this kind of agentic workflow could realistically fit into IT and business processes.

Alongside traditional workflow automation, I'm assessing how agentic automation could autonomously handle multi-step processes that currently need manual handoffs between systems — always with a human in the loop wherever judgment or approval matters.

Local LLM Lab

A hands-on lab for learning the tools before implementing them enterprise-wide. Runs on a three-node Proxmox cluster with centralized shared iSCSI storage, giving a realistic testbed for evaluating performance, resilience, and operational overhead at something close to production scale.

On top of the cluster, self-hosted local models via Ollama (Llama, Gemma, Mistral, Qwen) enable inference testing without sending private data to third-party APIs.

The goal is straightforward: know firsthand what on-premises AI can and can't do before it shows up in a vendor pitch.

Home Lab & Networking

Home infrastructure run the way an enterprise would run it — segmented, monitored, and recoverable. The core is a three-node Proxmox VE cluster with centralized shared iSCSI storage, hosting the VMs and containers behind everything else I experiment with.

Networking runs on Cisco Meraki and Ubiquiti gear. ZeroTier provides a secure mesh VPN for remote access without exposing ports to the internet, and Uptime Kuma keeps watch over services with alerting when something drops.

Home Assistant

A self-hosted Home Assistant deployment automating the home with an emphasis on local control and privacy — integrating devices from different vendors into one platform, with automations that don't depend on someone else's cloud staying up.

It's also a practical proving ground for ideas that matter at work: integrating systems that weren't designed to talk to each other, keeping data on-prem, and building automation people can actually rely on.

Obsidian PKM

An Obsidian-based personal knowledge management system — a linked, markdown-native vault covering technology decisions, project notes, and research. Everything is plain text: portable, versionable, searchable, and free of vendor lock-in.

Connected notes surface relationships between ideas that folders hide, and the same vault pairs naturally with AI tools for summarizing and cross-referencing what I've already learned.