Talent acquisition leader by day (linkedin.com/in/aaronboldt) — builder of AI-assisted applications on nights and weekends, working with Claude Code, Codex, and Claude Design in agentic development workflows.
Most of my product repositories are private because they're active commercial projects. This page describes what's in them. I'm always happy to walk through any of these live — the code, the architecture decisions, and the agent workflows that built them.
Stack: TypeScript · React Router · Prisma · Shopify App Bridge · Astro (marketing site)
A Shopify app that scans product descriptions and label images for the marketing claims that get supplement, CBD, and cosmetics stores in trouble — flagging risky language before the FDA, FTC, or a payment processor does.
- Hybrid detection engine: deterministic rules + LLM classification, with per-product category detection — supplement rules for supplements, cosmetics rules for cosmetics, automatically
- Cosmetics-specific checks including the appearance-vs-function boundary ("reduces wrinkles" is a drug claim; "reduces the appearance of wrinkles" is a cosmetic)
- Vision pipeline reads product label images for claims that never appear in the copy
- Category-aware guidance and one-click compliant rewrites for every flag
- Versioned engine with verdict caching and a calibrated benchmark suite in CI (0 false negatives / 0 false positives across the fixture catalog)
Status: Live on the Shopify App Store. Unit + integration test suites (dedicated Dockerized Postgres test DB)
Shopify App Store: apps.shopify.com/aplomb
Marketing site: aplomb.store
Stack: React 18 · TypeScript · Vite · Tailwind CSS · Supabase · Stripe · PWA
An installable progressive web app for mood journaling and habit tracking, designed around gentle, sustainable behavior change.
- Mood journaling and habit tracking with review flows
- AI-assisted insights (Anthropic / OpenAI, server-side via Supabase Edge Functions)
- Web push notifications, Stripe subscriptions
- Encrypted medication tracking — sensitive health data is encrypted at rest, and all secrets live server-side only
Status: In beta with early testers.
🎼 Impresari — a governed AI platform that helps founders validate, design, build, and launch software
Stack: Next.js · TypeScript · Tailwind CSS · Supabase (Postgres + Auth) · Vitest · Playwright
Impresari guides software projects from an initial idea through business viability, product definition, architecture, design, implementation, validation, and go-to-market planning. It combines specialized AI agents, verifiable project context, human approvals, and bounded execution to keep important decisions and actions under user control.
- Structured product intake covering customer problem, business viability, delivery capacity, application type, risks, and success criteria
- Guided phase dashboards, agent activation, traceable artifacts, decisions, evidence, and approval workflows
- Evidence-based ICP, positioning, channel, marketplace, launch-readiness, and post-MVP growth planning
- Layered architecture separating workflow policy, repository context, Portal authorization, and local execution
- Read-only GitHub integration with writes and other consequential actions separately gated
- Local Bridge/Runner security pilot for signed, bounded jobs in isolated proposal workspaces while reusable provider credentials remain locally controlled
- Deny-by-default activation, cancellation and revocation controls, sanitized evidence, and hosted validation at approved stages
Status: Active development and private preview. The local Bridge/Runner is a private, default-disabled security pilot; broader production agent execution remains gated
I start with PRDs to define what I’m building, who it’s for, and what matters most. I use ARDs to map out how major features should work and ADRs to record important technical decisions and why I made them. Claude Code and Codex support the work through structured planning, design, development, security, and testing stages. Each stage must pass reviews and automated tests before moving forward, and higher-risk features are tested safely before they can access real projects or take sensitive actions.

