Chattime

Inspiration

The project began with a simple but uncomfortable observation: AI is becoming so convenient that many of us skip the thinking step entirely. The brain is a muscle, and without the proper exercise… those muscles won’t grow to their full potential.

We paste emails, ask for rewrites, request summaries, and rely on AI for tasks we used to reason through ourselves. That raised deeper questions:

What cognitive tasks are we offloading?
How much thinking are we delegating?
At what point does convenience become cognitive dependency?

What it does

ChatTime analyzes a user’s prompt history and classifies each prompt into five cognitive categories drawn from Bloom’s taxonomy:

Repetitive
Information retrieval
Problem solving
Critical thinking
Creativity

It then groups these into AI Automation Tasks vs Human Cognitive Tasks, computes a Cognitive Offloading Index (COI), and generates a clear report showing:

How much thinking the user did
How much thinking the AI assistant did
Whether their usage pattern is healthy, moderate, high, or critically reliant
A pie-chart visualization of cognitive vs automated work
An insightful summary with recommendations

In short: ChatTime is a mirror for your AI habits.

How we built it

We started by grounding the system in educational psychology, especially Bloom’s taxonomy and research on cognitive offloading.

From there, we built the pipeline:

Prompt Parser — cleans and segments user prompt history
LLM-based Classification Engine — assigns each prompt to one or more cognitive categories
Metric Calculation Engine — computes category percentages and the COI using a weighted formula
Visualization Layer — generates the AI-vs-Human cognitive pie chart
Insight Generator — produces a short, human-readable usage summary

ChatTime accepts pasted prompt histories, processes them through this pipeline, and outputs a full AI Dependency Report.

Challenges we ran into

Defining cognitive categories: Human thinking is messy. Mapping prompts to Bloom-based categories required careful interpretation and iteration.
Weighting the COI: Choosing weights that reflect cognitive effort (e.g., creativity = 1.0, repetitive = 0.2) required balancing academic grounding with practical usability.
Ambiguous prompts: Many prompts span multiple cognitive levels (“design a scalable architecture” touches creativity, critical thinking, and problem solving).
Integrating the feedback system with Vertex AI through Google Cloud: One of our biggest technical challenges was getting the feedback system to work reliably with Vertex AI on Google Cloud. While we were eventually able to get the backend pipeline functioning, integrating that feedback layer smoothly into the frontend within hackathon time constraints was much harder, and the final front-end connection remained incomplete.

Accomplishments that we're proud of

Creating a quantifiable way to measure AI cognitive offloading (something rarely discussed but deeply important)
Designing a COI metric that is academically grounded yet intuitive for everyday users
Building a system that transforms raw prompt histories into meaningful insights and visualizations
Crafting a future-ready vision: a plugin that sits inside ChatGPT (or other commercially available AI tools) and analyzes sessions with one click

What we learned

AI usage isn’t just about productivity; it shapes how we think
People often underestimate how much cognitive work they outsource until they see it visualized
Bloom’s taxonomy is surprisingly powerful when adapted to modern AI interactions
Designing metrics around human cognition requires both technical precision and psychological nuance

What's next for ChatTime

Browser/ChatGPT plugin that automatically analyzes sessions without manual copy-paste
A fully integrated feedback system in the frontend so users can receive live, session-specific guidance
Feedback improvements using A/B testing to evaluate which recommendation styles are most effective for changing user behavior
Cosine similarity–based feedback matching to compare current prompt patterns against known cognitive usage profiles and generate more accurate, personalized feedback
Longitudinal tracking so users can see how their AI reliance evolves over weeks or months
Personalized recommendations to help users rebalance cognitive habits
Team/organization dashboards to understand collective AI usage patterns
Privacy-preserving on-device analysis so prompt histories never leave the user’s machine