AI Replica experiments
What if my portfolio could talk back?
Most portfolios are archives, a record of work that happened, waiting to be read. I've been experimenting with a different question: what if a portfolio could be present? Not just document who I am, but actually represent me, answer questions, hold a conversation, explain a decision, describe how I think?
These two experiments explore that question from opposite directions. One shows you a version of me. The other lets you talk to one. Neither is the answer. Both are the beginning of a question I think the design industry is about to have to take seriously: as AI makes it possible to scale presence beyond the physical, what do we actually want to preserve, and what are we willing to let a machine represent?
I built both in my own time, using my own expertise, as both the subject and the designer. That dual role taught me more about AI-human representation than any client project could.
EXPERIMENT 1
The Visual Replica: a version of me you can watch
The first experiment was the simplest version of the question: can AI faithfully represent my face, my voice, and my professional narrative in a format a stranger would find credible?
Using HeyGen, I trained a video replica of myself with my likeness and voice, producing a presentable AI version that can deliver a professional introduction without me being in the room. The production time was hours. The implications will take longer to work out.
What I learned: visual fidelity comes faster than most people expect. Within a short session, the replica could deliver my narrative convincingly enough that a recruiter or hiring manager encountering it cold would engage with it as they would a well-produced video introduction. The uncanny valley, that uncomfortable gap between almost-human and fully human, was smaller than I anticipated in professional presentation contexts.
What it can't do: it can only say what I scripted. It doesn't know me. It represents the surface of who I am: the words, the cadence, the face, without the depth. A visitor who asks a question gets silence. That limitation is the design problem the second experiment tries to solve.
EXPERIMENT 2
The Digital Mind: a version of me you can talk to
If HeyGen answered the question of presence visually, Delphi answered it conversationally. Where the replica can be watched, the Digital Mind can be questioned.
Delphi works by building what it calls a Digital Mind, an AI trained on everything you've created to represent your thinking. I fed it my portfolio case studies, my Medium articles, my career history, my design philosophy, and the frameworks I've developed across 12+ years of design work. The result is an interactive version of me that a visitor can call or chat with to ask about specific projects, my approach to trust design, how I'd approach a particular UX problem, or what I learned from a specific company.
The difference from a chatbot is significant: Delphi responds from my source material, grounded in what I've actually written and thought, not generic AI responses dressed in my name. It cites where answers come from. It has guardrails that prevent it from representing me in ways that contradict my documented thinking.
What it does well: it scales the conversation I can have with a hiring manager or collaborator beyond the constraints of my calendar. Someone who wants to go deeper than a portfolio case study can, at any time, without scheduling. For the senior roles I'm targeting, where a hiring manager might want to probe a candidate's design philosophy before committing to an interview, this changes what a portfolio can accomplish.
What it can't do: it can't replicate the human moments: the pause before a hard answer, the genuine curiosity, the relationship that builds over time. It's a starting point, not a substitute.
What building both taught me
① Presence and representation are different design problems. "HeyGen solves for presence: making me visually available at scale. Delphi solves for representation: making my thinking accessible at depth. A portfolio that combines both is something genuinely new: not a document you read, but an entity you can interact with. The design challenge is knowing which modality serves which moment."
② The source material is the design system. "Delphi's quality is entirely determined by the quality and coherence of what you feed it. My portfolio case studies, articles, and documented frameworks became the training data, meaning that the better I articulate my thinking in writing, the better my Digital Mind represents it. This is a new kind of design artifact: content built not just to communicate, but to train. The implications for how designers should document their work are significant."
③ The uncanny valley of expertise is harder to cross than the uncanny valley of appearance. "HeyGen crossed the visual uncanny valley faster than I expected: the face and voice were credible within hours. But the expertise uncanny valley, the gap between AI that sounds like it knows what I know and AI that actually does, is much harder to navigate. Delphi narrows it through grounding in the source material, but the gap remains visible under probing questions. That gap is the most interesting design problem in AI representation right now: how do you make expertise feel genuine when it's being synthesized rather than lived?
