Projects

Work I'm proud of.

A focused selection — built end to end, shipped with intention, and maintained with the same care I bring to customer work.

Featured project

RenewNudge℠

AI-powered Customer Success CRM

Open live app

A fully functional, solo-built multi-tenant SaaS CRM for renewal tracking, account health, and retention workflows — taken from concept to working product using AI-assisted development.

Role: Founder · Product & Engineering
Timeline: May 2026 — Present

ReactTypeScriptTanStack StartSupabaseTailwindPostgres RLS

The problem

Most CS teams discover renewal risk too late — buried in dashboards, Slack threads, and gut feel. The signals exist, but no one has time to stitch them together account by account.

Approach

Designed, built, and deployed a multi-tenant SaaS CRM with live auth, persistent data, and workspace-scoped accounts, renewals, risks, tasks, and playbooks.
Architected a secure backend with Postgres Row Level Security, role-based workspace membership, and SECURITY DEFINER policy helpers; ran security scans and resolved findings to eliminate cross-tenant data exposure.
Shipped end-to-end authentication with email/password and Google OAuth, team invites, and per-workspace seed and wipe flows — new users can sign up, invite teammates, and load sample data in under a minute.
Translated CS domain expertise across renewals, QBR prep, health scoring, and at-risk playbooks into a focused, action-oriented retention workflow.

Outcome

Live at renewnudge.lovable.app — designed so a CSM can go from a fresh signal to drafted outreach in under a minute. Pilot conversations in progress.

More work

Jun 2026

CanIRunLocalAI

Privacy-first local hardware scanner for AI model feasibility.

A local-only scanner that answers the question most teams actually have before adopting on-device AI: what models can this machine realistically run? Generates a hardware report and recommends practical local models based on RAM, VRAM, GPU vendor, and use case.

Approach

Built a cross-platform Python scanner that inspects RAM, VRAM, GPU vendor, and CPU capabilities without sending data off the device.
Mapped hardware profiles to a curated set of Ollama- and llama.cpp-compatible models with realistic quantization recommendations.
Shipped a clear, actionable report so a non-specialist can go from 'can I run this?' to a working local setup in minutes.

Outcome

A practical decision-support tool for teams exploring private, local AI — especially useful in regulated or privacy-sensitive customer environments where cloud inference is off the table.

PythonOllamallama.cppHardware Detection

Feb 2026 — Apr 2026

Personal AI Exploration Initiative

Self-hosted AI server for private chat, web search, persistent memory, and image generation.

A high-performance self-hosted AI server for local Stable Diffusion image generation, fine-tuning, and inference — showcasing hands-on GPU-accelerated computing on consumer hardware.

Approach

Configured and optimized Stable Diffusion environments (Automatic1111 WebUI, ComfyUI, Forge) using Docker and CUDA for efficient, low-latency generation on consumer-grade hardware.
Engineered the full server architecture end to end — hardware selection, OS install, driver configuration, and performance tuning to maximize VRAM utilization and minimize generation times.
Kept the surface area small and maintainable: one server, scoped access, easy to reason about.

Outcome

A dependable private AI environment for daily drafting, research, and image generation workloads — no external providers, no leaked context.

Stable DiffusionAutomatic1111ComfyUIForgeDockerCUDA

Jul 2026

SpotlightSaver

Local-only Windows utility for preserving Spotlight wallpapers.

A small, focused Windows utility that captures the current desktop or Windows Spotlight wallpaper before it rotates away, saving the image with matching local metadata — no uploads, no telemetry, no background service.

Approach

Wrote a minimal .NET 8 CLI that searches the standard wallpaper and Spotlight cache locations and copies the active image on demand.
Emits a companion metadata file with capture time, source path, dimensions, and file size — extracting local metadata where Windows exposes it.
Kept the tool local-only by design: no network calls, no telemetry, no persistent process.

Outcome

A dependable one-click save that respects user privacy — a small proof point for shipping focused, trustworthy tools.

C#.NET 8Windows