AI Enabled Portfolio
An AI-native portfolio site
This portfolio was designed by heavily leveraging an AI workflow for both development, and content refinement. What you see on the site is the design spec - there is no master design file - I have been iterating on the site design in code leveraging AI for quick cycle times.
I’ve also decided to run a little experiment and provide an AI interface for visitors of the site. It changes the portfolio from a passive document into something that literally answers the questions folks come to a portfolio to answer.
It’s definitely experimental, but times are changing and there’s too much fun to be had to not give it a whirl.
AI Workflow - Building tools to help deliver higher quality
I created an inline editing feature turns every content block into an editable field. A floating Haiku chat panel receives full context - what block I'm editing, the original text, what I've changed, the full case study content, and a master knowledge document. This is all to help ensure I'm getting feedback in real time rather than waiting for someone to take a look.
Because of the quality focus, the AI actually doesn't make the first draft faster - it slows it down. However, the first 'draft' ends up being much closer to ready because it acts as a perpetual critique partner. It catches buried leads, flags imbalanced depth, tells me where I'm over-indexing. The stuff a human reviewer would catch in a portfolio critique session, but available on every iteration. In a sense, it helps me "one shot" a portfolio piece.
Most of the copy on this site was workshopped here - not in a Google Doc, not in Figma. In the running page, discussing the content with an AI, in the website.
Making the site AI enabled
People come to a portfolio with questions, AI gives answers.
Every visitor comes to a portfolio trying to decide things like "Should I bring this person in?", "what are their strengths", "what's the scope of their work?", and more. Hiring managers for example, might need a systems thinker, or someone with exceptional visual design, or a rockstar in emerging tech. Traditionally portfolios make these folks infer the answer from case studies - hoping the right signal surfaces as they (often) skim.
The AI assistant flips that, and just attempts to meet the user need quickly. Visitors can ask directly: 'Is Austin a systems thinker? Show me examples.' Or 'What's his experience with AI product design?' And get targeted answers - including context that doesn't make the front page of a case study but is exactly what that specific person is looking for.
It's designed to make the portfolio serve a hiring manager's job-to-be-done, as quickly as possible.
How can visitors trust the bot?
The hardest part of designing the AI representing me on the portfolio has been making it honest. The default behavior of most LLMs is sycophancy - they'll hype you up, dodge weaknesses, and hallucinate accomplishments. That's worse than useless for a portfolio bot. If a hiring manager asks about a gap in my experience and the bot deflects, they'll trust nothing else it says.
The system prompt is tuned to give honest, critical answers. It's willing to say 'he's lighter on X' or 'that's not really his strongest area - here's where he's deepest.' The bot is designed to be credible, not flattering. Trust comes from honesty, not from hype.
The AI is designed to have more knowledge than is on the portfolio
The AI draws from a layered context system: a career narrative document with my full professional history, enrichment files with deep context on craft, design philosophy, leadership, and each project, plus the structured case study content from every portfolio page.
I built multiple interview and feedback tools to allow me to create extremely rich, structured context that allows the AI to present a much more textured and nuance insight into questions about me as a designer than you can really have in a standard portfolio presentation.
This means the bot can surface project details that don't earn a spot on the case study page but are exactly what someone might ask about - a specific technical decision, a stakeholder management story, a metric from a project three jobs ago. The portfolio is the curated surface; the AI has the full depth.
The context corpus is maintained alongside the portfolio itself. When I add a new case study, the AI's knowledge updates with it.
Designing the AI interface.
I wanted to have a simple call to action to chat available and prominent, but not make it so that you can't just browse the portfolio regularly. It's deliberately minimal until someone engages, and can always be toggled off.
When conversing with the bot, the page enters an alternate mode that allows users to add page elements as context by clicking a + icon next to cards, blocks, or text when it's highlighted The AI sees what the visitor is referencing, so its answers are grounded in the specific work the person is looking at.
This AI Enablement is a real risk and an experiment
There's real risk in building an honest AI representative. The bot is designed to give enable critical answers - which means it can genuinely undersell me if the prompt tuning isn't right, or surface a weakness out of proportion, or at an innoportune time. Every visitor interaction is a live demo with no safety net.
However, the industry is changing. Our roles are changing. Expectations are changing. I felt it was worth the risk to try something genuinely interesting and frankly, vulnerable on this site. Since I can't speak to much of my day-to-day AI work at Meta, projects like this allow me to show my approach to working with and on AI products in new and exciting ways.
The next thing I'd like to try is to generate the entire portfolio on the fly with AI, however that will take some deep thinking about the actual information and context architecture, and could be too experimental to support a proper portfolio visiting experience, so I'm starting simple with this rigorous AI enablement to see what the response is before going all in. Let me know what you think!