CareerPooler: AI-Powered Career Simulation

• CareerPooler is a generative AI-powered career navigation system that uses a pool-table simulation to model non-linear, effort-dependent career decision-making.
• The system employs a physics-based interface where milestone, skill, and random event balls interact dynamically, fostering experiential learning and reflective planning.
• Empirical studies show that CareerPooler outperforms linear chatbots in user engagement, satisfaction, and career clarity by emphasizing uncertainty and adaptive decision-making.

CareerPooler is a generative AI-powered system designed to reimagine career exploration through an interactive spatial-narrative simulation grounded in a pool-table metaphor. Unlike linear chatbots that generate deterministic, overly idealized guidance, CareerPooler presents career progress as a non-linear, effort-dependent process marked by uncertainty, setbacks, and serendipity. The system introduces a spatial and narrative framework where users enact career decisions by striking a cue ball, navigating major milestones, skill acquisitions, and unpredictable events—each dynamically generated by a fine-tuned LLM. This visually grounded, analogy-driven environment aims to foster deeper engagement, experiential learning, and a more resilient, practical understanding of career development.

1. System Architecture and Metaphor-Driven Simulation

CareerPooler operationalizes career decision-making as a simulated game of pool. The primary interface is a physics-based pool table where each "ball" corresponds to a specific event in the user's simulated career trajectory. Three principal entity types are encoded:

• Milestone Balls: Represent major career achievements or transitions (e.g., landing a job offer, obtaining a promotion, encountering a pivotal choice).
• Skill Balls: Denote professional skills or competencies acquired over time (e.g., leadership, technical proficiency, networking).
• Random Event Balls: Embody stochastic factors such as unexpected opportunities, challenges like funding loss, or neutral career events.

The user interacts by "striking" the cue ball, causing it to move and potentially collide with event balls. The collisions and resultant rebounds are governed by a physics engine, making the progression inherently non-deterministic and sensitive to user input. Hints, rendered as limited, contextually vague overlays upon hovering, allude to limited foresight in real-life career decisions.

Event generation is executed asynchronously: fine-tuned LLMs produce milestone narratives, skill descriptors, and event trajectories in context, leveraging user profile data and interaction history. The pipeline outputs are managed in modular JSON structures, ensuring logical consistency, chronological coherence, and contextual linkage.

2. Algorithmic and Interactional Mechanisms

The CareerPooler event-generation pipeline is orchestrated by a hybrid system:

• LLM Event Synthesis: LLMs synthesize events considering previous milestones, skills pocketed, and annotated user traits. Each narrative is tailored—milestones may account for prior successes or skill acquisition, while random events are generated with labels ([Positive], [Negative], [Neutral]) dictating their impact on progression.
• Collision-Driven Narrative Unfolding: System-level collision detection and response (frame-synced at high frequency) govern the onset of new events or consequences. A successful "shot" (e.g., the cue ball pocketing a milestone) may unlock new skills or trigger follow-on events, while misfires or rebounds simulate setbacks.
• Hint/Information Economy: Hints provided about future balls are intentionally cryptic, simulating information scarcity and risk in actual career paths. Users receive feedback through evolving event logs, timeline visualization, and animated feedback on the spatial interface.

The architecture is dual-panel: a physics-driven left panel for pool-table action, and a right panel that archives event logs and milestones in linear chronological order.

3. Empirical Evaluation and User Study

CareerPooler underwent a within-subjects experimental evaluation with 24 participants (including diverse student and early-career demographics). Each participant simulated a two-year career trajectory using CareerPooler and a linear chatbot baseline (ChatGPT, untuned for career guidance), with counterbalanced order.

Quantitative Metrics:

• Engagement: CareerPooler yielded significantly higher user engagement (absorption and immersion) than the chatbot baseline, $p \ll 0.01$ (as assessed by 5-point Likert scale surveys).
• Information Gain: Users reported that career information was more digestible and retained when interacting via the spatial-narrative interface.
• Satisfaction and Career Clarity: Overall satisfaction and career clarity scores were both statistically superior for CareerPooler over baseline; e.g., Wilcoxon signed-rank statistic $W = 12.0$, $p \ll 0.01$ for satisfaction.

Qualitative Insights:

User interviews revealed that the embodied interaction required participants to deliberate, sequence actions, and process failure constructively. The visible spatial arrangement and ambiguous prompts made failure, detour, and serendipity in career paths tangible—reportedly normalizing setbacks, increasing reflective time, and supporting experience-based learning. The metaphor enabled users to conceptualize career building as experimental and adaptive, rather than prescriptive.

4. Contributions to Career Guidance and Generative Systems

CareerPooler’s central innovation is its use of a non-linear, visually grounded, analogical interaction paradigm for career guidance. By departing from standard deterministic chat, it acknowledges that real-world career paths are seldom optimal, linear, or fully planned. The key system contributions are:

• Experiential Learning: Embodied spatial mechanics encourage users to learn from consequences, experiment, and recover from uncertainty, mirroring the complexities of actual career navigation.
• Engagement through Analogy: The pool-table interface, underpinned by real-time simulation and narrative hooks, demonstrably increases both engagement and emotional resonance.
• Information Economy and Clarification: By progressively revealing information and challenging the user with ambiguous hints, the system supports deeper cognitive processing and actionable career planning.

The approach signifies a broader methodological direction for generative AI systems in educational, career, and life-planning domains: shifting toward grounded, interactive simulations and metaphor-driven analogical environments as an alternative to linear text generation.

5. Technical Implementation and System Robustness

System performance is maintained through asynchronous, non-blocking event generation and physics rendering. The event pipeline is modular: each user interaction (e.g., shot, collision, pocket) is queued and processed such that narrative responses do not introduce latency into the interactive simulation. LLM-generated event content adheres to strict schema constraints, and event sequencing is managed within the game logic to ensure causally consistent and contextually meaningful progression.

Optimized collision and rendering code ensures high responsiveness, and decoupling of the event log/timeline display from the pool-table physics allows for simultaneous real-time interaction and narrative tracking.

6. Broader Implications and Prospects

The CareerPooler system contributes to the literature on AI-assisted career exploration by demonstrating that visually grounded, metaphorical interactions fundamentally alter how users approach and benefit from decision support systems. The evidence suggests that this class of systems can outperform LLM chatbots on critical user-centric metrics in career planning contexts. The spatial-narrative approach, by making uncertainty and path-dependence salient, has potential applications in other non-deterministic domains requiring resilience, such as financial planning, health behavior, and educational trajectory design.

These findings indicate that future career guidance platforms may benefit by merging generative AI with simulation, narrative, and interaction design, facilitating engagement and reflective learning in uncertain decision spaces.