PocketZot

Use AI productively with PocketZot!

Comment

Inspiration

Tamagotchi inspired web browser extension. Easy to use and entertaining.

What it does

PocketZot is a browser extension that monitors your prompts to AI and grades its degree of cognitive offloading.

Introduce the virtual pet concept: your pet's health reflects your prompting habits!

If you rely too heavily on AI → pet health drops. If you use AI thoughtfully → pet thrives.

How it works

Step-by-step Flow:

  1. User sends a prompt to any LLM
  2. PocketZot’s trained model evaluates the prompt on a scale (-3 to +2)
  3. Based on the score, your virtual pet gains or loses health
  4. Visual feedback reinforces better prompting habits

The Point System: Inspired by Bloom’s Taxonomy and broader research on cognitive offloading in AI-assisted learning, we developed our own grading framework and trained an LLM to evaluate prompts within a gamified system designed to encourage intentional AI use.

+2: The prompt limits the AI’s role to explanation or clarification while preserving the user’s responsibility for reasoning and solution-building.

+1: The prompt demonstrates active user reasoning and requests feedback or validation rather than full solutions.

-1: The prompt seeks explanation or information without significantly engaging the user in higher-order reasoning.

-2: The prompt outlines a task but relies on the AI to perform the central analytical or evaluative steps.

-3: The prompt transfers primary reasoning responsibility to the AI, asking it to generate conclusions or complete higher-order thinking on the user’s behalf.

The goal isn’t to punish AI use, but to encourage intentional usage.

How we built it

We built a web browser extension that integrates a fine-tuned large language model to automatically evaluate and classify user prompts based on our custom learning taxonomy. To power this system, we curated a labeled dataset and fine-tuned the OpenAI API to produce consistent classifications. We iteratively evaluated and refined the model’s outputs to improve reliability and alignment with our rubric. On the front end, we developed extension logic to monitor user prompts in real time, send them to our classification API, and process the returned scores. These scores are dynamically mapped to a gamified feedback system, where a Tamagotchi-style virtual pet’s health and behavior reflect the quality of the user’s prompts. We also designed the animation states and interaction logic to create an engaging feedback loop that encourages deeper, more thoughtful prompt construction.

Challenges we ran into

  • Getting the LLM to objectively and consistently rate user prompts

  • Needed to re-fine-tune the model after discovering it was classifying prompts in unintended ways

  • Significant scope creep that expanded beyond the original project plan

  • 32x32 px assets appeared blurry in the IDE due to compression/scaling issues

  • Accidental script changes within the folder (ex: copy-pasting images for ant population logic) caused breaking dependencies and instability

  • Learning curve with many unfamiliar technical concepts (including gravity logic and system mechanics)

Accomplishments that we're proud of

  • Designed and developed our first fully functional web browser extension

  • Built and optimized custom gravity mechanics and anteater roaming logic

  • Developed dynamic animations, including drag, throw, idle, and bounce states

  • Translated academic research into a working LLM-powered classification system

  • Bridged the gap between theoretical research and real-world application

What we learned

Throughout this project, we learned that behavioral design plays a critical role in shaping user interaction and motivation. We discovered that classification systems are far more complex than they initially appear, especially when striving for objectivity and consistency. Iteration proved to be unavoidable; refining our taxonomy, retraining the model, and adjusting system logic were necessary steps rather than setbacks. We also learned that managing scope is essential for survival in ambitious technical projects, as unchecked expansion can quickly derail progress. Most importantly, we recognized that AI is a powerful tool to augment our thinking, but it cannot replace human judgment, intentional design, and critical oversight.

What's next for PocketZot

  • Expand and clarify the taxonomy to better capture nuanced prompt behaviors

  • Fix persistent data storage issues to prevent resets on launch

  • Add more pet behaviors and interactive animations to deepen engagement

Built With

Share this project:

Updates