Technology

Building the future, one line of code at a time

Latest Tech News

Loading latest technology updates...

Projects I've Built with AI

Over the past year I've been building full-stack products end-to-end using agentic AI— GitHub Copilot agent mode paired with Claude to handle architecture, backend, frontend, infrastructure, and deployment. These are real, live products built from scratch.

🏠 homecardiy.com

● Live at www.homecardiy.com

Full-stack AI home & car inventory platform. Track every asset you own with photos, serial numbers, repair history, AI repair guidance, warranty alerts, and document storage.

  • Agentic repair engine: Claude uses tool calling to check your repair history, open recalls, and overdue maintenance before answering — autonomously, per query
  • 7 tools in the repair agent: repair history, safety recalls, maintenance status, YouTube, Amazon, local stores, contractors
  • Claude Vision API extracts appliance/vehicle details from photos automatically
  • 1,200+ unit tests, schema contract validation, staging gate before every prod deploy
  • AWS Lambda + Aurora DSQL + API Gateway + S3/CloudFront — built with GitHub Copilot agent mode

🗾 Japan Trip Companion

● Live at tlacy.github.io/tokyo-trip

Interactive travel app built for an 8-day Tokyo family trip (May 2026). Converts a static itinerary into a full mobile web app in a single HTML file.

  • Day-by-day itinerary with Google Maps & numbered stops
  • 70+ Japanese phrases across 7 categories with native speech synthesis (ja-JP)
  • Live speech-to-text translation via microphone + MyMemory API
  • All booking refs, emergency contacts, and transit tips in one place
  • Proof-of-concept for myitineraryai.com (see below)

🌐 myitineraryai.com

⏳ In Development — myitineraryai.com

Upload any trip itinerary (PDF, Word doc, or text) and get back an interactive mobile web app—personalized maps, phrases, tips, and booking refs—ready in minutes.

  • Claude extracts structured trip data from unstructured itinerary documents
  • Generates a bespoke trip companion app, hosted on S3
  • Stripe-gated: review your parsed itinerary first, then pay $7 to generate
  • Same Lambda + API Gateway stack as homecardiy.com

🔌 MCP Server (Professional Context)

● Try the interactive demo →

An MCP (Model Context Protocol) server that exposes my professional background as structured resources for AI assistants.

  • Career data, metrics, and writing available to Claude Desktop and other MCP clients
  • AI tools for bio generation, career queries, and engineering metrics
  • View on GitHub →

🍽️ weeklymealsai.com

● Live at www.weeklymealsai.com

AI weekly meal planner. Set your household size, diet, and budget—Claude generates a full week of meals plus a consolidated grocery list.

  • Claude generates structured meal plans from natural-language preferences
  • Auto-built grocery lists with reliable recipe search links
  • Same serverless stack: Lambda + Aurora DSQL + API Gateway + CloudFront

🌱 elawnaustin.com

● Live at www.elawnaustin.com

Multi-tenant field-service invoicing SaaS. eLawn Austin is tenant #1; the platform is built from day one to support any field-service contractor (lawn, sprinkler, HVAC, pool).

  • Strict multi-tenant data isolation—every query scoped by provider, cross-tenant tests required
  • Customer request → appointment → invoice → payment lifecycle
  • Reliability and security as the product—no AI, just rock-solid UX

📷 PhotoCraft

● Live at tomlacy.net/photocraft

RAW photo compositing platform. Upload NEF/DNG/CR3/ARW files and build HDR, focus-stack, best-blend, and panorama composites in the cloud.

  • Python Lambda containers running rawpy + OpenCV for real image science
  • Direct-to-S3 presigned uploads; async compositing with job-status polling
  • Solved real production hard problems: memory caps, linear-vs-sRGB gamma, OOM stale-job reapers

🚀 AI Zero to One

● Live at www.aizero2one.com

The platform behind every app on this page. It does two things that matter: it gets an AI agent to actually ship reliable software, and it runs for almost nothing.

  • Gets the best out of an AI agent—the human owns the outcome, the agent owns its craft and its mistakes, and forcing functions (tests-first, a staging gate, a schema contract) make the checks that matter unskippable. Every mistake becomes a documented pitfall the next session can't repeat.
  • Costs almost nothing to run—serverless top to bottom (Lambda + Aurora DSQL + CloudFront): usage-based pricing with a generous free tier that resets monthly, scaling to zero when idle. No fixed hosting floor.
  • One command from idea to deployed product—the full stack provisioned for you (Lambda, DSQL, API Gateway, S3/CloudFront, SES)—or take the source and own it outright.

AI Leadership & Approach

My philosophy on AI in engineering: the fastest way to learn what's real vs. hype is to build with it. That's what I've been doing.

🤖 Agentic AI Development

Building products using GitHub Copilot agent mode—autonomous multi-step coding that handles planning, implementation, testing, and deployment.

  • Agent writes tests first, catches its own errors, iterates without prompting
  • Full-stack apps (DB schema → API → frontend → CI/CD) in days, not weeks
  • Deep familiarity with where agents excel and where human judgment is irreplaceable
  • Documenting pitfalls and patterns in 100+ documented agent instructions per project—a living playbook that gets tighter every session

👁️ Multimodal AI (Vision)

Production use of Claude Vision API to extract structured data from photos, documents, and insurance cards—no forms required.

  • Photo → appliance details (brand, model, serial number) automatically
  • Insurance card → policy number, carrier, vehicles extracted in one call
  • Client-side resize before upload to stay within Lambda's 6MB invocation limit
  • HEIC handling, format validation, graceful degradation for unsupported types

☁️ Serverless AI Infrastructure

Deploying AI-powered APIs on AWS serverless stack built for minimal ops overhead and cost efficiency at startup scale.

  • Lambda (Node.js) + API Gateway HTTP v2 + Aurora DSQL (IAM-auth PostgreSQL)
  • S3 + CloudFront for global frontend delivery
  • AWS Secrets Manager for zero-plaintext credential management
  • Staging gate in deploy pipeline: smoke tests run against staging before production

⚡ Engineering Productivity at Scale

At the enterprise level: establishing AI-assisted development practices that improve velocity without sacrificing quality or security.

  • AI-assisted code review, test generation, and documentation
  • DORA metrics to measure the actual impact of AI tooling on delivery
  • Security-first AI adoption: OWASP compliance, prompt injection defense
  • Build vs. buy framework for AI capabilities (when to use APIs vs. fine-tune)

Engineering Leadership Principles

  • Build to learn: The fastest way to have an informed AI opinion is to ship something with it
  • Agents need guardrails, not leashes: Agentic AI is most powerful with tight feedback loops (tests, contracts, staging)
  • Parity over convenience: Dev/prod environment gaps cause more production bugs than bad code
  • Scalable architecture: Design for 10x growth, optimize for today's constraints
  • Security first, and re-checked: zero-trust, defense in depth, and OWASP by default—plus automated dependency and secret scanning across every project and a human-owned security review on a cadence, because AI-era risk doesn't hold still

Lessons Learned Building Six Products with AI

  • The model is the easy part: the hard part is the infrastructure, security, and test discipline around it
  • Capture every pitfall: a documented mistake becomes a guardrail—the agent instruction files now prevent classes of bugs across every new project
  • Test before browser: a 0.2s unit test beats a 10-minute deploy-and-check loop every time
  • Own the mistake, fix it, write it down: Christopher Avery's Responsibility Process applies to humans and agents alike
  • Rules need an owner: guardrails and tests only produce reliable software on an ownership model—the human owns the outcome, the agent owns its craft, and forcing functions make the essentials unskippable. It draws on Avery's Responsibility Process and Geoff Woods' CRIT.
  • Security has to keep pace with AI: the machine layer—dependency and secret scanning across every repo, on a schedule—is table stakes; the part that actually protects you is human-owned: triaging alerts, rotating secrets, auditing access on a cadence. Automation flags; a person owns the outcome.
  • Reuse the engine: each product feeds lessons back into a shared starter, so the next build starts further ahead

Influences & Staying Current

I keep this practice sharp by learning from the people pushing AI-augmented leadership and engineering forward:

  • Geoff Woods — The AI-Driven Leader (and his work on AI leadership): using AI as a thought partner for better executive decisions, not just a productivity tool
  • Gene Kim — Vibe Coding: building software through intent and conversation with AI rather than line-by-line authoring
  • Plus the daily feed above—I read the news, but I learn by shipping