Hi, I'm Andy.

I work frontline in Newegg's PC Support — troubleshooting hardware, answering compatibility questions, handling returns, and helping customers choose the right parts. Talking to customers all day, I kept hitting the same friction that isn't obvious from a pure web-design view: people unsure where to even start for help, technical information scattered across product listings, support pages, and manufacturer docs, and assisted sales that were never credited back to the agent who guided them. So I approached it as both a support rep and a UX designer — turning real support patterns into a connected set of design concepts, prototypes, and proposed workflows for internal review and demos.

Before, there was no clear place to send customers for PC help and the information was scattered across product listings, support pages, and manufacturer docs. The concept pulls everything into one hub with several clear entry points:

Iteration: it wasn't one-and-done. An early concept put a "Buy Now" button directly in chat; after stakeholder feedback I removed it and shifted to the lighter-touch attribution approach — recognizing the agent's contribution without pressuring the customer. The result is a clearer, less intimidating support journey: defined entry points, hardware concepts anyone can grasp, a structured callback experience, a simpler trade-in flow, and a real way to credit agent-assisted sales — all drawn from patterns in actual customer cases. The goal was never to replace agents, but to clear up confusion before the conversation starts so they can focus on the hard troubleshooting.

Work shown here consists of designs, prototypes, mockups, and internal demonstration materials — not necessarily deployed to Newegg's live production site.

STUBOT started as a joke: type “STUBERT” in Discord and it posts a random photo of my dog, Stuart. That tiny keyword bot became my hands-on way to learn Python — and I never stopped expanding it. It grew commands, a personality, and permissions; became a remote 3D-printer monitor with status, progress, ETA, and live webcam snapshots over Discord; gained real conversation by integrating the OpenAI API; and finally moved off my computer onto a dedicated Raspberry Pi 5 so it runs around the clock without a paid cloud server. Each new thing I learned became another layer.

I wanted more structure in my days without adding another app I'd end up ignoring. Paper is hard to ignore, so I built on what I'd learned self-hosting STUBOT: a Raspberry Pi 5, this time wired to a printer. Every morning at 6:30 it prints a one-page sheet called Seven Steps Towards Convergence: a mantra and a short true story to start the day fired up, six fresh tasks pulled from different parts of my life, a Victory Lap reward at the end, and one personal question for the night. Before bed I check in from my phone, tick off what I finished, and answer the question by typing or just talking. Epoch folds my answers into a private memory and writes the next morning's sheet around them, so every sheet fits my life a little better than the last.

How I built it: I ran it like a product, with myself as the first user. The hardware came first: moving the Pi to boot from an NVMe drive, getting an older HP inkjet printing over USB with driverless IPP, and serving the check-in page over HTTPS with the Pi's own certificate authority so my iPhone would allow the microphone. Python ties it all together. Claude returns structured JSON that has to pass validation before anything prints, WeasyPrint turns HTML and CSS into the printed page, and systemd timers run the schedule, all for a few dollars a month in API costs. Then I iterated from daily use. When the steps started to feel repetitive, I made them new every day. When typing on my phone felt like friction, I added voice to every field. When I wanted more motivation, the morning mantra became a coach that talks to me like a winner. Everything lives on the Pi; the only data that leaves the house is what's sent to the AI APIs.

AI as a power tool: I built Epoch with Claude as my AI pair programmer. I owned the product: the vision, the requirements, and the decisions about how it looks, feels and behaves. I set up and debugged the hardware, tested each change on real paper, and backed the code with automated tests that simulate the AI, the microphone and the printer, plus browser tests on a phone-sized screen.

What it shows: When I see a real need, I build the whole thing: hardware, software, design, and the unglamorous reliability work that lets it run every single morning without me touching it. I design for the person actually using it, and I keep improving from honest feedback, including my own.

Once Epoch was running, I wanted the same Pi to do something for someone I love. The Cheli Chronicle is a weekly newspaper made for an audience of one. The front page opens with an over-the-top “Hear ye, hear ye!” proclamation declaring her week the best one imaginable, packed with the things she loves: otters, penguins, turtles, chinchillas, space, math and science. Stuart, our chihuahua, weighs in from Stuart's Corner as Head of Household Security. There's a fun fact, the real moon phase for the week, and a big picture to color. The back has a crossword, a word search, and a secret message where every letter is a small math problem, with the answers printed upside down at the bottom.

How I built it: Claude writes the words, but the puzzles come from custom code, not the AI, so they're always solvable: a crossword engine that fits the week's answers into an interlocking grid and numbers the clues, a word-search generator, and a cipher that turns a sweet message into simple math. Each puzzle is checked against its answer key before it can print. The coloring page comes from OpenAI's image model and is cleaned into high-contrast line art with Pillow, with automatic fallbacks if a model is unavailable. Two-sided layout, Saturday scheduling, a one-tap test print and a web page for my notes all plug into the same Epoch system. Like Epoch, I built it with Claude as my AI pair programmer.

What it shows: Technology is at its best when it's personal. I care about the details nobody asked for, like making sure every crossword is solvable and every moon phase is right, and I like using what I know to make the people around me smile.

The issue shown is a sample that uses one of the built-in drawings. Real issues get a new AI-drawn coloring page every week.

Panda Security Solutions is my own small-business concept — residential security-camera installation, computer troubleshooting, and custom PC builds. I rebuilt its website from a basic page into a modern, professional platform: a clean black-and-white panda brand, a clear service layout, and a customer journey that ends in a quote form. Behind the scenes it runs fully serverless on AWS, so submissions are captured and stored without a traditional web server ever running. The hard part wasn't the visual redesign — it was modernizing the front end without breaking the live back end that processes real customer leads.

From a basic page to the live site — rebranded, rebuilt, and deployed by utilizing AWS S3, CloudFront, API Gateway, Lambda, DynamoDB & CloudWatch.

Before

After