Reset

Inspiration

What it does

Inspiration

Students are constantly told what they should do to feel better: sleep more, use their phones less, take breaks, manage their workload, and exercise.

But there is a fundamental problem: what works for one person may not work for another.

We wanted to build something that went beyond generic wellness advice. That became Reset — a personal wellness experimentation platform that helps students discover what actually works for them.

The idea was inspired by a simple question:

What if, instead of telling students what should work, we helped them test it?

What We Built

Reset follows a simple loop:

Track → Discover → Hypothesize → Experiment → Measure → Learn

Students can log daily factors such as stress, focus, energy, sleep, workload, screen time, socializing, and context switching. Over time, Reset can identify potential patterns in that data.

But we deliberately don't treat correlations as facts.

For example, Reset might observe:

“Your overwhelm has tended to be higher on days with more workload and context switching.”

Rather than claiming that one caused the other, Reset turns the observation into a hypothesis that the student can test.

Students can create N-of-1 experiments comparing interventions such as Cognitive Offload and Single-Task Sprint. Episodes are randomly assigned to intervention arms, outcomes are measured afterward, and the results are analyzed using permutation testing.

This lets Reset answer a much more useful question:

“Which strategy appears to work better for you?”

rather than:

“Which strategy is supposed to work?”

Over multiple experiments, the results contribute to a Personal Playbook — an evidence base of strategies that have, or have not, worked for that individual.

The Importance of Uncertainty

One of our core design principles is that Reset is allowed to say “we don't know yet.”

If an experiment does not have enough data, Reset does not manufacture a conclusion. Instead, it explicitly reports insufficient evidence and encourages the user to continue collecting observations.

We believe this is especially important in a world where AI systems are often expected to provide an answer to everything.

Reset's philosophy is:

If the evidence isn't there, don't pretend it is.

How We Built It

We built Reset as a full-stack web application with a frontend focused on making complex experimentation feel approachable and intuitive.

The core experiment engine handles:

• Randomized intervention assignment
• Intervention-before-outcome timing
• Baseline separation from experiment data
• Dropout handling
• Two-sided permutation testing
• Exact permutation enumeration when feasible
• Sampled permutations for larger experiments
• Minimum evidence requirements before drawing conclusions

The statistical engine was developed separately from the interface so that the UI cannot simply “make up” an interpretation of the data. The experiment results are generated from the underlying trial methodology and then presented in language that students can understand.

We also built a Find Support feature using the CMS NPPES NPI Registry to help users find listed healthcare providers, while deliberately avoiding claims that an NPI listing means a provider is currently licensed or credentialed.

What We Learned

Our biggest lesson was that personalization is not the same thing as personalization through AI.

It would have been easy to build a chatbot that looked at a student's data and generated personalized wellness advice. But that still leaves the fundamental question unanswered:

Did the advice actually help?

Building Reset forced us to think about the difference between an observation, a hypothesis, and evidence.

We also learned how difficult it is to communicate statistical uncertainty without making an application feel like a statistics textbook. Concepts like randomization and permutation testing are powerful, but students shouldn't need to understand the mathematics to use the product.

That led us to design Reset around simple conclusions such as Strong Evidence, Some Evidence, and Still Learning, while keeping the underlying methodology available for deeper inspection.

Challenges We Faced

One of our biggest challenges was designing an experiment system that was statistically meaningful while still being simple enough for a student to use.

Small implementation details could completely change the validity of an experiment. We specifically had to address issues involving the timing of interventions and outcomes, as well as ensuring that the set of permutations used during analysis matched the experimental design.

We also had to resist the temptation to add features simply because they sounded impressive. A wellness app can quickly become a collection of mood tracking, meditation, AI chat, reminders, recommendations, and dashboards.

Instead, we kept returning to one question:

Does this help a student discover what actually works for them?

That became the filter for our product decisions.

What's Next

Reset is designed to become more useful as a student continues using it.

More observations lead to better hypotheses. More experiments lead to a stronger Personal Playbook. And experiments that don't produce meaningful evidence remain part of the user's learning process rather than being discarded.

Our long-term vision is simple:

Reset shouldn't tell everyone the same thing. It should help each person build their own evidence about what works.

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Reset

Built With

• react
• vscode