The idea was straightforward.
We tend to enjoy doing things we're good at. But the abilities that come most naturally to us can be the easiest to overlook. When something has always felt easy, we may never recognize it as a strength at all.
Vocari began with the idea that conversational AI could uncover those overlooked patterns, then connect them to careers where those abilities could become an advantage.
Academic research prototype with a limited career catalog and early-stage personalization. The link demonstrates current capability, not the full product vision.
The idea started with something I didn't know about myself.
I had always thought of myself as a creative person. If someone had described me as a “systems thinker,” I probably would have pictured structure, repetition, math, and rigid processes. It wasn't a strength I would have chosen from a list because I didn't know it described me.
After nearly two years of conversations with ChatGPT across work, school, projects, problems, and ideas, it pointed out a pattern: I was consistently thinking in systems.
More importantly, it explained what that meant using evidence from my own experiences.
Suddenly, things I had thought were unrelated started to connect. My creativity and systems thinking weren't opposites. Seeing connections, noticing what wasn't working, understanding how one change affected everything around it, and imagining a better way for the pieces to work together were part of the same pattern.
Having language for that ability changed the way I understood myself. It expanded the kinds of problems I believed I could solve and the kinds of work I could imagine myself doing.
And it left me with a question:
Why should an insight like that take two years to surface?
What if an AI experience were intentionally designed to look across someone's experiences, find patterns they may not recognize themselves, and help them understand what those patterns could mean?
Where could those abilities take you?
That question became Vocari.
The concept was simple.Building it was not.
Vocari started as a passion project, then became the focus of a group project in an Experience Design course. I kept owning and building the prototype while my team and I used the course to research, test, and challenge the idea. The build was anything but easy, but what I learned from each problem was irreplaceable. Every limitation exposed something new to understand, and each lesson shaped what I built next.
Before the course began, I had already created a very basic version of Vocari in Lovable. My team and I decided to build our Experience Design project around that concept. They contributed to the research and testing while I continued owning the prototype itself.
The affective-computing work produced useful signals, but the aggregated output made it difficult to connect a response to the exact prototype moment that may have triggered it.
As the prototype grew, Profile DNA, patterns, discovery, reactions, and recommendations were all describing the same person in slightly different ways. I stopped adding screens and defined the concepts underneath them.
Once I could trace how evidence became a recommendation, I realized much of the adaptive experience was still being produced by explicit rules rather than an AI model interpreting the person's context.
A user rejecting Product Designer should not blindly remove a card or rewrite their entire profile. The reason for the rejection determines what, if anything, should ripple into other careers.
Product and design careers were overrepresented, and a career match still did not tell someone how to get from their current background to that career.
Dashboard
Vocari is an unfinished working prototype under active development. I could start another product and repeat the parts I already know: research, flows, interfaces, prototyping, testing, and iteration. Instead, I am continuing with Vocari because the questions left are the ones I want to learn from.
Right now that means implementing AI-assisted interpretation for human-in-the-loop feedback, refining the rules that control how evidence ripples through the system, broadening and standardizing career coverage, and defining how a person's background can connect a career match to realistic education and training pathways.
Requirements, constraints, rules, structure, calibration, and safeguards are not cleanup after the product. They are the product work I am doing now.
Persistent evidence, Profile DNA, Progressive Discovery, structured reactions, career exploration, recommendation logic, and an expanding AI interpretation layer.
The current prototype is not production-ready, and I do not present unresolved pathway, calibration, career-coverage, or safeguard questions as solved.
Lovable → GitHub → ChatGPT → Supabase → Vercel → testing
When prototyping limits interrupted iteration, I moved relevant codebase context into ChatGPT for targeted changes, worked through GitHub version control, and used deployment workflows to keep the prototype available. That workaround also pushed me deeper into the architecture behind my product decisions.
The idea was straightforward.
Building it wasn't.
That's why I'm still building it.
Information architecture and search strategy redesign for a global open-source software marketplace.