SenceBank

An interactive Unity game that uses Gemini AI to analyze your in-game choices and movements, offering fun, personality insights you can share with friends—designed to be accessible for all people.

Comment

Inspiration

We love personality tests, but long text surveys aren’t friendly for busy people, kids, or anyone who prefers visual, low-reading experiences. We asked: what if everyday choices could be banked into a quick “sense snapshot” instead? Morning chaos felt perfect—universal, playful, and time-boxed.

What it does

SenseBank is a choice-driven mini-game about how you start your day. You wake up at 7:00 AM and aim to leave home by 10:00. In each scene, make real-life choices — breakfast, outfit, packing, or maybe a few distractions. Every decision shapes your time, mood, and personality traits. At the end, you’ll receive a short, friendly personality readout and 2–3 practical tips for your daily life.

There’s no right or wrong — just play naturally and see what your morning says about you.

How we built it

Scenes in Unity to WebGL, embedded in React. Unity streams a compact event log (action IDs, tags, time/mood deltas, timestamps) to a serverless function, which calls Google Gemini for a schema-shaped reply at low temperature; a local scorer covers outages. We kept the UI light: mono type, soft gradients, tuned routing/caching, so WebGL loads fast on laptop and phone.

Challenges we ran into

  • Captured Unity scene events and passed structured data to the web app

  • End-to-end AI summary that feels personal without being clinical

  • Cute, OO UI that people wanted to try again and share

Accomplishments that we're proud of

  • End-to-end AI summary that feels personal without being clinical.

  • A lightweight schema-validated pipeline that’s robust to flaky networks.

  • Cute, clean presentation that people wanted to replay and share.

  • Gathering all the information and finishing everything on time.

What we learned

For most team members, it was their first time using Unity. We learned how to manage scene flow, use prefabs, handle simple state management, and build a complete WebGL project from start to finish.

Schema-first prompting > free text. Designing a JSON response schema (with low temperature) gave us consistent, render-ready summaries.

Timeboxing matters. Coding time disappears faster than you think!

What's next for SenceBack

More scenes & smarter flow: New places, times, and situations + AI-assisted dynamic branching that generates follow-up choices based on what the player already picked.

Localization & accessibility: Reduced-text mode, high-contrast theme, full keyboard navigation, and screen-reader labels.

Embeddable widget for real flows: A lightweight module that can be embedded in third-party journeys (e.g., bank-account sign-up or checkout). With explicit consent, we turn in-flow choices into gentle personalization and product recommendations. If we explore credit-related use cases, it will be opt-in research only with clear disclosures and include compliance & fairness reviews.

Deeper personality analytics: More traits/facets, clustering and trend tracking across sessions, exportable result cards, and a small API for dashboards and A/B testing.

Share this project:

Updates