PersonaPath

Inspiration

When most people picture their future career, what comes to mind is rarely the actual day-to-day of the job. It's the title, the salary range, the thing your family tells you to do, and how it sounds when someone asks, "so what do you do?" Those things are the easiest to picture, which is why they're the things we chase.

But none of them tell you whether you'll walk into your workplace energized in the morning, or silently worn down by it at the end of the year. Both of us kept landing on the same question: not "is this a good career?", but "is this a good career for me? Does it call on the things I'm naturally strong at, or constantly demand the things I struggle at? Will the work actually feel rewarding, or will it just tire me out over time?"

That's the uncertainty we wanted answered going into our future. It’s also the one PersonaPath is built to navigate.

What it does

PersonaPath starts where most career tools stop by taking note of a user’s personality. It extracts traits using familiar frameworks such as MBTI, then allows users to select potential careers, mapped to O*NET occupational codes. These codes contain structured, day-to-day descriptions of what a role actually demands, including factors such as task repetitiveness, level of independence, frequency of interpersonal interaction, and broader work style and environmental context.

PersonaPath then performs a structured comparison between a user’s traits and O*NET occupational dimensions using deterministic mappings. An LLM is used only as an explanation layer, translating pre-computed results into clear, human-readable insights rather than generating conclusions directly. This ensures that all interpretations are grounded in computed compatibility signals rather than model inference.

The system highlights where a user’s personality is structurally aligned with a career, as well as where meaningful friction exists. Instead of focusing only on fit, it surfaces tradeoffs that are typically invisible from job titles alone, and distinguishes between friction that can be managed through adaptation and friction that is inherent to the nature of the role.

When multiple careers are selected, PersonaPath presents them side by side for direct comparison. From there, users can choose to pursue a path as-is, pursue it with adaptations, or continue exploring additional options.

Once a decision is made, the system generates an actionable roadmap that translates the chosen career path into concrete next steps, bridging the gap between decision-making and execution.

Ultimately, PersonaPath is a decision-support system that helps users understand the implications of different career paths, rather than a “best career” recommendation engine.

How we built it

Stack

Frontend: Next.js, React, TailwindCSS
Backend: FastAPI (Python)
Data: O*NET Web Services API

Instead of just letting an LLM map personality to each occupational reality directly through inference, we built a hard-coded rule set that checks the user’s traits against O*NET to calculate alignment and friction, providing determinism. Gemini API then enters afterward to turn those calculated results into something readable, like an explanation, a strategy, or a roadmap. Everything was built inside Google Antigravity IDE, an agent-first platform that handled the implementation once we settled the architecture ourselves.

Challenges we ran into

Grounding LLM explanations into reality

The earliest challenge was to prevent the AI from making personality-career connections that sound plausible, but are not supported by evidence. We want every conclusion to be traceable to actual occupational data rather than allowing the model to infer everything entirely. To solve this, we used O*NET data as the foundation for Gemini to explain and synthesize from, providing citations whenever used.

Missing occupational data

O*NET data isn’t equally detailed across every occupation. Psychologists, the first career we evaluated, had no data at all. To address this, we introduced a confidence system that displays the density of available data, rather than presenting every result with the same level of certainty.

Generating meaningful and actionable explanations with AI

Generating useful responses was a difficult problem. Early on, most outputs simply relayed abstract explanations and advice that were not very actionable. We revised our prompts heavily to ensure the AI provided descriptions and guidance that are tangible, concrete, and meaningful.

Avoiding reducing users to personality labels

Ensuring the system did not treat personality frameworks as absolute definitions of a person was one of the most important challenges we faced. MBTI and Enneagram are useful tools for identifying behavioral tendencies and recurring patterns, but they cannot fully capture an individual. These frameworks are also often considered pseudoscientific, and should not be used to prescribe advice as if they were diagnoses.

Early iterations risked framing users as static archetypes, producing explanations that implied a personality type defined the user. We redesigned the reasoning process to treat profiles as groups of patterns that resonate across many people.

The system now reasons: “If this pattern resonates with you, then this friction or tradeoff may be relevant.”

This shift preserved the value of personality-informed analysis while allowing users to decide which insights actually reflect their lived experience.

Accomplishments that we’re proud of

Building a decision-support system for a question that mattered to us

Both of us, along with many of our peers, spent a lot of time asking the same question: “What do I want to do in the future?” Choosing a career is one of the most significant decisions a person makes, yet it is often approached with very little understanding of the day-to-day reality of the work.

We’re proud to have built a system that tackles a problem that genuinely mattered to us. Taking an idea rooted in personal uncertainty and turning it into a tool that could help others navigate the same question was one of the most rewarding parts of the project.

Grounding AI in real occupational data

Using our rule set with O*NET occupational data allowed us to keep AI-generated explanations transparent and grounded in real-world evidence. Rather than relying on model inference, every insight is associated with structured data from an authoritative dataset. This ensures PersonaPath remains accountable to the information behind its conclusions.

Deploying our first live web application

This was the first time either of us built a project that became a publicly accessible web application. Deploying an end-to-end system for real users was a major milestone.

What we learned

AI is most effective when grounded in structure

One of the biggest lessons came from initially allowing the AI model to directly infer career compatibility from personality data. This produced plausible-sounding but ultimately ungrounded assumptions. By introducing O*NET-based deterministic rules, we made outputs more transparent and justifiable, and significantly improved reliability.

Understanding career friction is more useful than recommendations

We found that users gained more value from understanding tradeoffs and friction in career paths than from direct recommendations. This shifted the system from a “decision engine” to a “decision-support tool.”

Re-framing personality frameworks as patterns

Another key lesson was how our understanding of personality frameworks like MBTI and Enneagram evolved. Initially, they felt reductive when applied to real people. However, as we integrated them into our system, we came to appreciate their value more deeply.

Rather than treating them as fixed labels, we found them useful as structured patterns in how people think and work. When combined with occupational data, they surfaced meaningful signals about alignment and friction.

In other words, building PersonaPath helped us realize the value of personality frameworks beyond pop psychology. Their practical value lies in exploring how cognitive and behavioral patterns interact with real-world contexts, rather than labeling individuals.

What’s next for PersonaPath

Moving forward, we plan to expand beyond MBTI and Enneagram by incorporating more empirically grounded models such as the Big Five and HEXACO. Unlike typology-based systems, these frameworks represent personality as continuous traits rather than fixed categories. This will allow us to move toward richer, more dimensional and academically grounded user profiles.

Alongside this, we plan to introduce structured modeling of concrete interests and hobbies. While personality traits describe cognitive tendencies, interests provide clearer signals about what people are naturally drawn to in practice.

By combining trait-based models with interest-based signals, we can build a more complete representation of both cognitive style and motivational direction.

These improvements will strengthen PersonaPath’s ability to reason about career compatibility using research-backed personality models and a more grounded understanding of what people actually enjoy doing.