MirandaLLM
A conversational layer for my portfolio. I designed, built for trust, and evaluated it end to end. (2026)
Launched
2026
My role
Solo
Research
Interaction design
Prompt design
Front-end
Tools
Claude API
Vercel
React
Framer
Product strategy
Building LLM products under a conditional output logic.
Recruiters and hiring managers are often busy and spend little time reviewing a portfolio to find the best candidate for a role, and the questions that matter most, like whether a designer has shipped production code or how they handle research, are usually buried several pages deep.
I designed and built MirandaLLM, a conversational layer for my portfolio that answers those questions in my own words, so reviewers can find what matters to them in seconds instead of minutes.
The central challenge wasn’t building a chatbot but making one people could trust. I set strict guardrails: never guess, never fill gaps, and never say “I don’t know,” to avoid misleading my large language model and to build trust.
Visual principles
It should feel like part of my portfolio, not a plugin
As I was designing my first LLM assistant for my portfolio to answer all my work-related questions. My first thought was the operational logic, like a language picker, so I built the letters in a spherical shape inspired by the display in the Asian Art Museum, and made it look like a library for picking words or sentences to match my portfolio's sphere visual style. I've also added shortcuts for common questions in the open chat window to allow users to quickly get involved in the conversation and steer people to find the answers they might care most about.
Inspired by the SF Asian Art Museum
States are a communication problem
The sphere doubles as a status indicator: it turns slowly at rest, wakes up and grows when you hover or open the chat, and spins with rapidly changing letters while an answer is being written. Instead of a generic loading spinner, the motion tells visitors exactly what the assistant is doing, and it stays completely still for anyone who has reduced motion turned on.
Still
Rest
Hover
Thinking
Contrast is a constraint problem
My portfolio moves between light and dark themes, so the Ask button couldn’t rely on a fixed color. It would disappear on one background or another. I designed it to read what’s behind it and switch between black and white, always choosing the option with the stronger accessibility contrast, including over images and mid-grey tones. For the button itself, I designed a small sphere of letters, a visual metaphor for a language model choosing words.
To explore the MirandaLLM, you can try it on the home page