Case Studies
Four examples of real, shipped work — not mockups.
Internal Operations Platform — 150-Store Car Wash Chain
Problem: A 150-location car wash company ran its employee lifecycle — hiring, onboarding, scheduling, training — across disconnected SaaS tools (JazzHR, ADP, Deputy, TalentLMS) with no shared source of truth, creating manual re-entry work and inconsistent records store-to-store.
Approach: Designed and built an internal intranet platform as the connective layer between those systems, automating the joiner/mover/leaver lifecycle instead of leaving it to manual cross-referencing. Sole author of the platform's core (300+ commits), plus a separate backend service handling cron-based job orchestration, third-party integrations, and an AWS migration with security hardening along the way.
Outcome: Store managers and HR got a single system reflecting reality across all 150 locations instead of four disconnected tools, with the error-prone manual sync work eliminated.
Production Database Access Fix — Firebird / DynDNS Root Cause
Problem: A car wash operator's point-of-sale system database (Firebird, on-site server) had become unreachable for ad-hoc reporting. Initial theories — wrong driver, network encryption, IP allowlisting — all turned out to be dead ends.
Approach: Traced the actual root cause: the server's hostname was a DynDNS address whose IP had silently rotated, and tooling elsewhere in the stack had a stale cached IP baked in instead of resolving the hostname live. Fixed the immediate connection (matching the client driver version to the server's actual Firebird version), then built a small, permission-scoped CLI reporting tool so future queries wouldn't need direct database exposure.
Outcome: Reporting access restored, and the stale-IP failure mode — likely to recur given the DynDNS setup — documented so it's a five-minute fix next time instead of a multi-day investigation.
Lazlo's List — Subscriber Product Built and Shipped Solo
Problem: Build a real subscription product from nothing: a weekly newsletter of free offers and giveaways, with a free tier and a paid early-access tier, on a tight budget (shared hosting, no dedicated infrastructure spend).
Approach: Built the full stack solo — an automated research pipeline that finds and vets candidate items, a PHP membership backend (anonymous/free/paid tiers, authentication, sessions), a Friday-early/Tuesday-public timing model enforced by a gated endpoint rather than a database flag, and a subscriber email pipeline with unsubscribe handling. Deployed on shared hosting with zero incremental infrastructure cost, with GitHub Actions handling deploy-on-push.
Outcome: A live, functioning subscription product — real signups, real paid-tier gating, real weekly sends — built and shipped end-to-end by one person.
Custom AI Voice Agent with Purpose-Built Tool Integrations
Problem: Off-the-shelf voice assistants can talk, but they can't take action inside a specific business's actual systems without custom integration work.
Approach: Built a voice-driven AI agent on top of Claude, with a custom MCP (Model Context Protocol) server layer giving the agent real tools to call — not just conversation, but actions grounded in real data and real systems.
Outcome: A working demo agent that holds a natural voice conversation and executes real actions through its custom tool layer, proving the pattern for turning an LLM into something that does the work instead of just describing how to do it.