Alvin Terrible

Engineering Lead | AI Subject Matter Expert

Featured Projects

Agentic Engineering Toolkit for Claude Code

Creator and maintainer · 2024 - present

Claude Code, Model Context Protocol, Node.js, TypeScript, PowerShell, GitHub Actions

A personal, project-aware toolkit that turns AI coding assistants into a governed part of the delivery pipeline. 46 slash commands, 42 skills, 17 lifecycle hooks and 68 automation scripts cover spec-driven development, automated pull request review, CI preflight checks, release documentation and security gates, steered per project by a descriptor registry so one toolkit serves every repository.

  • Turns ad-hoc AI usage into a repeatable engineering workflow: specify, implement, preflight, review, ship.
  • Governed by 32 architecture decision records, including a credential-gate hook that blocks secret exfiltration at the tool boundary.
  • Includes a self-improving instruction loop: failures captured in a queue are promoted into permanent rules only when they recur.
AI ToolingDeveloper ExperienceAutomationGovernance
Agentic Engineering Toolkit for Claude Code

me-brain - Local-First RAG Memory for AI Assistants

Architect and developer · 2025 - present

TypeScript, Node.js, LanceDB, Ollama (bge-m3 embeddings), Model Context Protocol, Azure Cosmos DB

A private retrieval-augmented generation system that gives AI assistants grounded, personal context. A Node/TypeScript CLI and MCP stdio server over a LanceDB vector store, embedded locally by Ollama so nothing leaves the machine. Separate corpora for facts, personal voice and third-party messages are never blended, and every ingestion path runs through a redaction gate with quarantine for high-confidence secret hits.

  • Ten MCP tools expose retrieval to any MCP-capable assistant without shipping data to a third-party service.
  • Idempotent ingesters with watermarking pull from work trackers, pull requests and chat exports; re-runs are no-ops.
  • Mines recurring workflows from real transcripts to suggest new automation, closing the loop with the toolkit above.
RAGVector SearchEmbeddingsMCPPrivacy by Design
me-brain - Local-First RAG Memory for AI Assistants

Command Centre Web - Enterprise Access Control Client

Engineering Lead (Frontend) · 2024 - present

TypeScript, React, Redux Toolkit / RTK Query, Nx, Tailwind CSS, Vite, Vitest, Playwright

Frontend delivery lead on the browser client for Gallagher's physical access control platform, used to run security operations for governments, airports, hospitals and critical infrastructure. Work spans feature delivery in a large monorepo, accessibility to WCAG 2.1 AA, internationalisation through Lingui and Crowdin, performance budgets and end-to-end coverage in Playwright.

  • Leads specification, review and release for frontend work in a regulated, security-critical product.
  • Raised the review bar with automated checks that run before a pull request is opened, not after CI fails.
  • Modernised legacy screens into a component-driven React architecture without disrupting operators.
ReactEnterpriseAccessibilitySecurityMonorepo
Command Centre Web - Enterprise Access Control Client

AI Enablement Programme - AI-Assisted Engineering Adoption

AI Subject Matter Expert · 2024 - present

GitHub Copilot, Claude Code, Model Context Protocol, GitHub Actions, Azure DevOps

The people side of AI-assisted engineering: standards, guardrails and coaching that let a team use generative AI on production code without eroding quality. Covers instruction and skill authoring, prompt and context design, automated AI code review, secret-handling policy, and the review practice that decides what AI-generated code is allowed to ship.

  • Coaches and mentors engineers one to one and in review, from first prompt to agentic workflows.
  • Wrote the always-on instruction layer and scoped rules the team's AI assistants read in the repository.
  • Treats context as a budget: measures what the always-on instruction payload costs and prunes it on a schedule.
Generative AIEnablementMentoringCode Quality
AI Enablement Programme - AI-Assisted Engineering Adoption

AI Data Categorisation Platform

Project lead · 2023 - 2024

ML.NET, OpenAI API, C#, .NET Core, Azure Functions, Azure Service Bus, Cosmos DB

Led delivery of a data categorisation platform for an Australian client that combined a trained ML.NET model with the OpenAI API, so cheap deterministic classification handled the bulk and the language model handled the ambiguous tail. Integrated with Azure Managed Applications and Service Bus for high-throughput processing.

  • Automated a manual categorisation process across large data volumes.
  • Event-driven Azure architecture kept throughput high and costs predictable.
  • Delivered as a managed application the client could run in their own tenancy.
Machine LearningOpenAIAzureEvent-Driven
AI Data Categorisation Platform

Programmatic Video Generation Pipeline

Project lead · 2023 - 2024

React, Remotion, TypeScript, C#, .NET Core, Azure

Led a project that generated video programmatically from React components using the Remotion library, turning structured content into rendered video without a manual editing step. Engineered a rendering pipeline that handled dynamic, data-driven scenes at scale.

  • Replaced hand-editing with a repeatable, code-defined rendering pipeline.
  • Component-based scenes made new video formats a code change, not a new production.
  • Scaled rendering throughput for batch generation.
ReactRemotionMediaAutomation
Programmatic Video Generation Pipeline

alvin.terrible.nz - Edge-Rendered Portfolio

Designer and developer · 2025

React, TypeScript, Vite, Tailwind CSS v4, Hono, Cloudflare Workers, Vitest

This site. A React single-page app prerendered to static HTML at build time and served from Cloudflare's edge as one deploy unit, with a Hono API owning the contact endpoint. Migrated off Azure Static Web Apps and .NET Functions; structured data, a generated sitemap and an llms.txt summary make it legible to search engines and AI crawlers that never run JavaScript.

  • Static-first rendering means crawlers and AI answer engines see real content, not an empty root div.
  • Public endpoints are rate-limited and capped so an unauthenticated form cannot flood the inbox.
  • Typecheck, lint, unit tests and the prerender build all gate every pull request in CI.
Cloudflare WorkersSSGSEOEdge Computing
alvin.terrible.nz - Edge-Rendered Portfolio