Full-stack developer & designer
ElizabethStein.
I design and build software that’s actually running in production. Eighty-six things shipped. Here are the eight that matter, live right now.
The Problem
Managed-service providers (MSPs) and healthcare entities lacked an automated, auditable methodology to quantify organizational cybersecurity readiness. Critical assessments previously relied on manual spreadsheets and disconnected consultations.
System Architecture
Architected a unified 12-phase assessment platform spanning a governed Dataverse relational schema, Power Automate automated scoring orchestration, and an externalized Next.js 16 + React 19 assessor web app with Microsoft Entra ID (MSAL) role-based access control.
Production Outcome
Live in production for a national cybersecurity nonprofit. Sole developer owning data models, business logic, front-end architecture, and deployment pipelines.
Context & Scale
Managing rapid client delivery across a diverse portfolio of agency sites spanning service, e-commerce, and high-traffic editorial content.
Technical Execution
Built and maintained 11 production Craft CMS sites utilizing PHP, Twig templates, Composer plugin pipelines, Herd local parity, and Craft Project-Config CLI for deterministic schema migrations.
Engineering Tooling
Developed internal MCP server wrappers and QA diagnostics (rocket-vitals) that automated regression scans and content-model validations across client deployments.
The Concept
Input any date from 1958 to 2020 and receive an authentic, era-aware cultural snapshot (Billboard #1, box office hits, economic FRED baselines, and historical events) synthesized into a cohesive time-travel narrative.
Algolia Multi-Index Pipeline
Dispatches parallel queries across four Algolia indices (Billboard, TMDB, FRED prices, Wikimedia). An LLM synthesizes the structured payloads into an era-specific narrative rendered inside a custom CRT/VHS scanline interface with Upstash rate-limiting and Langfuse observability.
Contest Victory
Won the Algolia Agent Studio Challenge ($750 USD + DEV++ award) against a global field of developer submissions, praised for making deep data archives intuitive and engaging.
Core Capability
Analyzes codebases via TypeScript AST parsing to narrate architecture, dependencies, dead code, and complexity hotspots in first person across 12 distinct persona modes.
Model Context Protocol
Exposes 14 Model Context Protocol (MCP) server tools over stdio and SSE, allowing Claude Desktop and cursor agents to explore codebase graphs with structured AST precision rather than raw grep queries.
Distribution & Tests
Published on the public npm registry as @purplegumdropz/specter with 65 CLI commands, 216 automated tests, and cross-platform Node.js/Bun support.
Grounded Vision AI
Unlike generic screenshot-to-code generators that hallucinate non-existent markup, Trace enforces an in-prompt whitelist of shadcn design system primitives using Google Gemini 1.5 Pro.
Live Sandpack & Self-Repair
Renders the generated TypeScript JSX live in an editable in-browser Sandpack sandbox, with an automated compile-check repair loop and real-time axe-core accessibility auditing with one-click fixes.
Contest Winner
Winner of the DEV.to GitHub Finish-Up-A-Thon competition. Features 38 automated tests and live confidence inspection.
Hybrid RAG Pipeline
Combines Pinecone vector embeddings with BM25 keyword retrieval and Tree-sitter AST parsing to generate and maintain accurate technical code documentation.
Automated Git Sync
Listens to GitHub webhook events to trigger background documentation rebuilds on git push, eliminating stale developer docs.
SaaS Monetization
Full subscription billing with Stripe ($35-$239/mo across Pro, Team, and Business tiers), Supabase authentication, and Redis job queues.
DevOps Aggregator
A high-performance command-line operations tool built on Bun and TypeScript that queries 10 service APIs in parallel (GitHub, Vercel, Sentry, Stripe, Notion, ClickUp, UptimeRobot, Railway, Jira).
Actionable Signal Filtering
Aggregates state across 90+ repositories into prioritized work contexts (CRC / Rocketpark / Personal), filtering out routine noise to surface only broken builds, failing webhooks, and blocking reviews.
Reliability
216 unit and integration tests ensuring reliable daily operation as the primary terminal dashboard for all production projects.
Realtime Voice AI
Integrates the OpenAI Realtime voice API for low-latency conversational mock interviews with real-time speech evaluation and dynamic follow-up questioning.
FSRS-5 Scheduling
Utilizes the state-of-the-art Free Spaced Repetition Scheduler (FSRS-5) algorithm to schedule review intervals across 1,300+ interview questions categorized by company tracks (FAANG + Stripe + Uber).
Production SaaS
Live commercial web application with Stripe subscriptions, Supabase persistence, and 150 automated tests.