StudyTransit

Find your flow. AI-powered study space discovery and focus sessions for students.

Created by Noel Negron & Zaryaab Khan

Built for TreeHacks 2026.

---

Inspiration

We kept seeing students wander between libraries, check crowdedness, and lose focus before actually studying. We wanted to cut that friction: a single app to discover the right space, lock in with a focus timer, and track progress with streaks and exam prep. StudyTransit brings discovery, focus, and accountability into one flow, so you spend less time hunting and more time studying.

---

What It Does

StudyTransit helps Columbia University students discover the right study spaces, stay focused with a timer, and track progress with streaks and exam prep—all in one app.

Page	Description
Home	Browse Columbia libraries with live occupancy data. See "Best bet now" (least crowded library). Rate spaces (1–5) to contribute crowd-sourced data.
Discover	Set preferences: intensity (deep/steady/social), noise level (silent/busy/buzz), group size, session length. Describe needs in plain text. AI recommends 3–5 spaces with personalized reasons.
Lock In	Pick task type (exam/reading/writing/project), energy level, and your current space. Run a focus timer (10 sec–3 min). Rate your focus after each session and get AI coaching tips.
Stats	Weekly dashboard: study streak, focused minutes, completed sessions. Set weekly goals. Add exams manually or upload a syllabus PDF to extract dates and study topics. Get AI insights and exam-specific tips.

---

How We Built It

React + Vite SPA with React Router. Libraries and study spaces are stored in Firebase Firestore (collection Libraries with subcollection spaces per library); the app falls back to built-in Columbia data if Firebase is not configured. Sessions, exams, ratings, and weekly goals live in LocalStorage. Claude (Anthropic) is preferred for AI; OpenAI is fallback. In dev, Vite proxies /api/anthropic to avoid CORS. PDF parsing uses pdfjs-dist, then Claude extracts exam dates and study topics. Discover uses getAISpaceRecommendations for AI-curated space matches (with rule-based demo fallback when no API key).

---

Challenges We Ran Into

• Syllabus parsing – PDFs vary a lot. We used text extraction plus Claude to get exam dates and topics, with fallback "recommended" topics when parsing failed.
• Claude from the browser – CORS and direct-browser access required some setup. We iterated on prompts and models until AI responses were consistent.
• Scope – Balancing AI features (recommendations, syllabus parsing, insights, tips) with a simple UI. We cut or simplified features to keep the core flow clear.

---

Accomplishments That We're Proud Of

• End-to-end flow – Discovery → Lock In → Stats, all working together.
• AI integration – Multiple AI touchpoints (recommendations, syllabus parsing, study tips, weekly insights) with fallbacks when the API key is missing.
• Exam workflow – Upload a PDF, get exams and study to-dos, and check them off as you go.
• Clean, themed UI – Columbia University branding with consistent footer nav and modals.

---

What We Learned

• AI works well for structured tasks (parsing syllabi, summarizing stats, generating tips) when prompts are clear and outputs are constrained.
• LocalStorage is enough for a demo; sessions, exams, and ratings persist without a backend.
• Small UX choices (shorter demo timer, simpler reflection, no clutter) make the app feel calmer and easier to use.

---

What's Next for StudyTransit

• Campus integration – Real library occupancy or space availability from Columbia systems.
• Backend + auth – Sync sessions and exams across devices and users.
• Virtual study rooms – Zoom integration for one-click group study links.
• Mobile app – Native experience for on-the-go discovery and focus.

---

Tracks We Are Applying To

We are in the Education track (sponsor: Zoom). TreeHacks 2026 Education Grand Prize awards the top study/learning apps that help students discover spaces, focus, and track progress.

StudyTransit fits the Education track by:

• Helping students find the right study spaces (libraries, quiet rooms, group areas)
• Lock In focus timer for productive study sessions
• Stats dashboard with streaks, exam deadlines, and AI-powered study tips

---

Tech Stack

• Frontend: React 18, Vite 6, React Router 6
• Data: Firebase Firestore (libraries, spaces, capacity)
• AI: Anthropic Claude (primary), OpenAI (fallback). Demo mode when no API key.
• PDF: pdfjs-dist for syllabus text extraction + AI for exam/topic parsing
• Charts: Recharts for space occupancy trends
• Persistence: LocalStorage for ratings, focus sessions, exams, goals

---

Project Structure

react-app/
├── src/
│   ├── App.jsx              # Routes + AppShell with footer nav
│   ├── main.jsx
│   ├── theme.css             # Columbia University theme (primary blues)
│   ├── styles.css
│   │
│   ├── pages/
│   │   ├── HomePage.jsx      # Library list, best bet, rating modal
│   │   ├── LibraryPage.jsx   # Spaces for one library (SpaceCard grid)
│   │   ├── DiscoverPage.jsx  # Preference form → AI recommendations
│   │   ├── LockPage.jsx     # Focus timer, reflection, AI tips
│   │   └── StatsPage.jsx    # Streaks, goals, exams, syllabus upload, AI insights
│   │
│   ├── components/
│   │   ├── SpaceCard.jsx    # Occupancy display, weekly chart, community avg
│   │   ├── FooterNav.jsx    # Home / Discover / Lock In / Stats
│   │   └── Logo.jsx
│   │
│   ├── state/
│   │   └── AppState.jsx     # Global state (libraries, spaces, ratings, sessions, exams)
│   │
│   ├── services/
│   │   ├── aiClient.js      # Claude/OpenAI: recommendations, coaching, syllabus parsing
│   │   ├── pdfParser.js     # Extract text from PDFs
│   │   └── firebase.js      # Firestore init (alternative to firebase/)
│   │
│   ├── firebase/
│   │   ├── firebase.js      # Firebase app + Firestore db
│   │   ├── firebase_utility.jsx  # fetchAllLibraries, fetchAllSpacesFromLibrary, updateSpaceCapacityAndCounter
│   │   └── seed_columbia_libraries.js.js  # Admin seed script (Libraries collection)
│   │
│   └── hooks/
│       └── useRecommender.js  # Rule-based scoring (noise, intensity, capacity)
│
├── scripts/
│   └── seed-firestore.js    # Prints flat libraries/spaces JSON (different schema)
├── vercel.json              # SPA rewrites for Vercel deploy
└── vite.config.js           # Proxy /api/anthropic → Anthropic API (keeps key server-side in dev)

---

How It Works

Libraries & Spaces (Firebase)

• Collection: Libraries (doc IDs: lib1, lib2, …)
• Fields: library_name, optional location
• Subcollection: Libraries/{libraryId}/spaces — each space has room_data:
  • space_name, space_capacity, space_counter
  • Day keys: Monday, Tuesday, … for weekly occupancy trends

The app loads libraries on mount, then fetches spaces per library. If Firebase is not configured, it falls back to hardcoded Columbia data in AppState.jsx.

Occupancy & Ratings

• space_capacity / space_counter → ratio (capacity ÷ people rated) → fullness 1–5
• When a user starts a Lock In session and rates a space, the app increments space_counter and adds the rating to space_capacity (weighted). Firestore is updated via updateSpaceCapacityAndCounter.
• Community ratings (1–5) are stored in LocalStorage and used for Discover scoring and SpaceCard display.

AI Features (aiClient.js)

Function	Purpose
getAISpaceRecommendations	Matches spaces to preferences; returns JSON array with space, library, reason
getSessionCoachingSummary	Post-session tip based on task, energy, duration
getWeeklyStudyInsights	Interpret weekly stats into actionable suggestions
parseSyllabusExams	Extract exam/midterm dates from PDF text
extractSyllabusTopics	Extract study topics from syllabus for an exam
getExamStudyTip	One contextual tip for upcoming exam
getFocusPrompt	Short pre-session motivation

Uses Anthropic first; falls back to OpenAI if no Anthropic key. Demo mode returns built-in responses when neither key is set.

Focus Sessions (Lock In)

• Sessions are stored in LocalStorage (lionstudy_focus_sessions).
• Each session: started_at, duration_minutes, task, energy, completed, reflection (optional focus rating).
• Stats page computes: weekly minutes, study streak (consecutive days with ≥1 completed session), task breakdown, rule-based coaching tips.

Exam Workflow

• Add exams manually (name + date) or upload a syllabus PDF.
• PDF → extractTextFromPdf (pdfjs-dist) → parseSyllabusExams + extractSyllabusTopics (AI).
• Each exam has a checkable topic list; "Get a study tip" calls getExamStudyTip.

---

Run Locally

cd react-app
npm install
npm run dev

Open http://localhost:5173

---

Environment Variables

Create react-app/.env:

Firebase (required for live library data)

VITE_FIREBASE_API_KEY=your-api-key
VITE_FIREBASE_AUTH_DOMAIN=your-project.firebaseapp.com
VITE_FIREBASE_PROJECT_ID=your-project-id

Without these, the app uses built-in Columbia libraries/spaces (no Firestore).

AI (optional – fallback to demo responses)

# Anthropic (Claude) – checked first
VITE_ANTHROPIC_API_KEY=sk-ant-api03-your-key

# OpenAI – used when Anthropic key is not set
VITE_OPENAI_API_KEY=sk-your-key

In dev, Vite proxies /api/anthropic to avoid CORS and keeps the API key on the server side.

---

Seeding Firestore

To populate Firestore with Columbia libraries and spaces:

cd react-app
node src/firebase/seed_columbia_libraries.js.js /path/to/serviceAccountKey.json

Requires a Firebase service account key. Writes to Libraries collection with library_name and spaces subcollection containing room_data (space_name, space_capacity, space_counter, day counts).

---

Demo for Judges

1. Home – Browse libraries, rate a space (pick space + crowdedness 1–5).
2. Discover – Pick intensity, noise, group size, duration. Click "Find me a spot" for AI-curated recommendations.
3. Lock In – Select task, energy, and your space. Start the timer (10 sec for quick demo). After reflection, click "Get a tip for next time" for AI advice.
4. Stats – Complete a few Lock In sessions first, then "Ask AI for weekly insights." Add an exam manually or upload a syllabus PDF. Open an exam to see the study to-do and "Get a study tip."

---

Deploy

Configured for Vercel (vercel.json):

cd react-app
npm run build