Mirror72

About the Mirror72

Most health and wellness tools focus on the past. They collect data, summarize behavior, and assign scores after decisions have already been made. The real problem is not understanding what happened, but understanding what is going to happen next. Mirror72 was built to address that gap. Mirror72 is a behavioral modeling system that takes in daily lifestyle inputs such as sleep, alcohol use, smoking, stress, hydration, and workload, and predicts how those inputs will influence a user’s mood and energy over the next seventy two hours. The system also allows users to simulate changes in behavior and observe how those changes alter their projected future state. Instead of simply reporting data, Mirror72 provides a forward looking model of human behavior.

Inspiration

The idea came from observing how often people experience delayed consequences without recognizing the cause. Poor sleep, substance use, and high stress rarely affect someone immediately. The impact is usually felt the next day or even several days later. As students balancing academic work, jobs, and personal life, we have all experienced sudden drops in energy, mood instability, and lack of focus without clearly understanding why. Over time, it became clear that these outcomes follow patterns that can be modeled.

Existing tools such as WHOOP and Apple Health provide useful tracking, but they stop at visualization. They do not attempt to model short term future states or allow users to test alternative decisions.

That gap led to the development of Mirror72.

How We Built It

Mirror72 is structured as a lightweight behavioral prediction system with a clear separation between data input, state modeling, and explanation.

Frontend The interface is built using React with Vite for fast development. Tailwind CSS is used for layout and styling. Recharts is used to visualize the seventy two hour forecast as a time series.

Backend The backend is implemented using FastAPI in Python. It exposes endpoints that accept user inputs, process them through the modeling engine, and return structured predictions.

Core Modeling Engine

The system converts user inputs into internal state variables that represent the user’s condition at a given moment. These include:

Energy Mood Recovery Stress load Crash probability

Each variable is computed using weighted relationships derived from intuitive behavioral effects.

For example, a simplified fatigue model can be expressed as:

Fatigue= (8−Sleep)⋅2+Stress⋅1.5+Alcohol⋅1.2Fatigue=(8−Sleep)⋅2+Stress⋅1.5+Alcohol⋅1.2

These state variables are then projected forward over time using simple decay and recovery functions. Recovery improves with rest and hydration, while stress and poor habits increase strain.

A simplified projection can be written as:

Future Energy=Current Energy+Recovery Influence−Stress DriftFuture Energy=Current Energy+Recovery Influence−Stress Drift

This allows the system to generate a continuous forecast over a seventy two hour window.

Simulation Layer

The simulation layer is the core of the system. Users can modify inputs such as sleep duration, hydration level, or alcohol consumption and immediately see how those changes affect the projected outcome. This is implemented by recomputing the state variables and forecast in real time. This transforms the system from a passive tracker into an interactive decision engine.

AI Explanation Layer An AI model is used to generate natural language explanations of the predictions. It receives structured output from the modeling engine and translates it into clear insights.

For example:

“Your projected energy drop is driven by low sleep and high stress. Increasing sleep tonight will have the largest positive effect on your recovery.”

The AI does not generate predictions. It explains them.

Challenges

One challenge was balancing realism with simplicity. Human behavior is complex, but the system needed to remain interpretable and responsive. The focus was on modeling strong and intuitive relationships rather than attempting full biological accuracy.

Scope was also a concern. The idea of a full digital twin is too broad for a short project. The system was intentionally limited to a seventy two hour prediction window to keep it focused and buildable.

What We Learned

Short term prediction is often more actionable than long term forecasting. Users respond more strongly to outcomes that are immediate and understandable.

Separating modeling logic from explanation improves both clarity and trust in the system. Designing around decision making rather than data collection leads to a more useful product. Users want to know what to do next, not just what they did before.

Future Potential

Mirror72 can be extended by integrating with wearable devices to automate data collection. It can learn from user history to personalize predictions over time. It can expand into areas such as burnout detection, mental health support, and performance optimization. More broadly, it can evolve into a personal behavioral intelligence layer that helps users understand and optimize their daily decisions.

Closing

Mirror72 represents a shift in how personal health data is used. Instead of tracking the past, it models the near future. It allows users to see the consequences of their actions before they happen and adjust accordingly. At its core, Mirror72 answers a simple question: What is going to happen if I keep living like this?