Track 2: Educational Equity | 7th Annual Datathon
EquiPath is an AI-powered, equity-centered college recommendation system that helps low-income and underserved students find institutions where they're most likely to succeed and graduate. Unlike traditional college search tools that prioritize prestige or rankings, EquiPath uses data-driven personalized scoring to center the needs of first-generation students, student-parents, and students from historically marginalized communities.
The Problem: Students from low-income backgrounds graduate at significantly lower rates than their higher-income peers—a gap of over 15 percentage points on average.
Our Solution: EquiPath matches students with colleges using a personalized Student Success & Equity Score that adapts to each student's unique circumstances, prioritizing affordability, equity outcomes, and student success over prestige.
EquiPath offers four ways to find your perfect college match:
- 🔍 College Finder - Quick form-based search with instant recommendations
- 💬 AI Chat Assistant - Conversational interface powered by Claude AI that explains recommendations in plain language
- 🗺️ Interactive Map - Geographic exploration with 6,800+ schools, filterable by state and your recommendations
- 📊 Data Insights - Visualizations showing the income-graduation gap that motivates EquiPath
Our proprietary algorithm combines four key dimensions:
- 💰 Affordability - Net price, affordability gap, childcare costs (for student-parents)
- 📈 ROI - Median earnings 10 years post-graduation vs. median debt
- ⚖️ Equity - Race-specific graduation rates and equity parity across demographics
- 🚪 Access - Admission selectivity and student GPA fit
The scoring weights automatically adjust based on your profile:
- Low-income students: Affordability weight increases by 15%
- Student-parents: Affordability +10%, Equity +5%
- First-gen/marginalized students: Equity weight increases by 10%
- Zip Code Radius Search: Find colleges within a specific distance from your location
- State Filtering: Focus on in-state schools for lower tuition
- Interactive Map Visualization: See all schools geographically with detailed popups
Colleges are clustered into meaningful archetypes based on their characteristics:
- Equity Champions (5.5%) - High ROI, high equity, affordable
- Balanced Options (29.4%) - Well-rounded across all dimensions
- Equity-Focused Access (20.1%) - Strong support for diverse students
- Good Value Options (10.8%) - Balanced affordability and outcomes
- Balanced Options 2 (34.3%) - Alternative balanced profile
- Uses official federal datasets (IPEDS, College Scorecard, Affordability Gap data)
- All scoring logic is transparent and auditable
- AI (Claude) used only for explanations and chat, NOT for decision-making
- Algorithm is deterministic and reproducible
-
College Results View 2021 (6,289 institutions)
- Graduation rates by race/ethnicity
- Median earnings and debt
- Admission rates
- Institution characteristics
-
Affordability Gap AY2022-23 (21,299 records)
- Net price data by family income bracket
- Affordability gaps (standard students & student-parents)
- Childcare cost adjustments
- State-specific minimum wage data
-
Schools GeoJSON (6,812 schools)
- Geographic coordinates for map visualization
- Institution names and addresses
Merge Strategy: Institutions are joined on UNITID for maximum accuracy. We filter to the $30,000 earnings ceiling (lowest income bracket) to ensure one record per institution. Final dataset: 4,933 institutions with 288 features.
7th Datathon/
├── app.py # Main Streamlit landing page
├── requirements.txt # Python dependencies
├── README.md # This file
├── TECHNICAL_DOCUMENTATION.md # Detailed technical docs
├── data/
│ ├── College Results View 2021 Data Dump for Export.xlsx
│ ├── Affordability Gap Data AY2022-23 2.17.25.xlsx
│ └── .cache/ # Parquet cache files (auto-generated)
├── map/
│ ├── schools.geojson # 6,812 school locations
│ └── index.html # Original HTML map (deprecated)
├── tableau_exports/ # CSV files for Tableau visualizations
│ ├── income_graduation_data.csv
│ ├── income_categories_summary.csv
│ ├── key_statistics.csv
│ └── TABLEAU_GUIDE.md
├── src/
│ ├── data_loading.py # Load and merge datasets
│ ├── feature_engineering.py # Compute institution-level metrics
│ ├── user_profile.py # UserProfile schema
│ ├── scoring.py # Personalized scoring logic
│ ├── clustering.py # K-means clustering for archetypes
│ ├── llm_integration.py # Claude AI integration
│ ├── distance_utils.py # Zip code distance filtering
│ └── config.py # Configuration constants
└── pages/ # Streamlit multi-page app
├── 1_🔍_College_Finder.py # Form-based finder
├── 2_💬_AI_Chat_Assistant.py # AI chat interface
├── 3_🗺️_School_Map.py # Interactive map
└── 4_📊_Data_Insights.py # Income/equity visualizations
- Python 3.11+
- Pandas & NumPy - Data manipulation
- scikit-learn - K-means clustering, StandardScaler
- Streamlit - Web application framework
- Plotly - Interactive visualizations (Data Insights page)
- Folium - Interactive maps
- Anthropic Claude API - AI chat assistant
- PyArrow - Parquet caching for fast data loading
- pgeocode & uszipcode - Geographic distance filtering
# Navigate to the project directory
cd "7th Datathon"
# Install dependencies
pip install -r requirements.txt
# Set up your Claude API key (for AI Chat Assistant)
# Create a .env file in the project root:
echo "ANTHROPIC_API_KEY=your_api_key_here" > .env# Option 1: Standard way
streamlit run app.py
# Option 2: If streamlit not in PATH
python -m streamlit run app.pyThen open your browser to http://localhost:8501
- Navigate to 🔍 College Finder in the sidebar
- Fill out your profile:
- GPA (0.0-4.0)
- Annual budget
- Home state
- Zip code (optional - enables distance filtering)
- Search radius in miles (optional)
- Race/ethnicity
- Income bracket
- First-generation status
- Student-parent status
- Preferences (public only, in-state only, school size)
- Click "Find My Matches"
- Review your top 10 personalized recommendations with:
- Match scores (0-1, higher is better)
- Distance from your zip code (if provided)
- Net price, graduation rates, earnings data
- Detailed breakdowns across all 4 scoring dimensions
- Navigate to 💬 AI Chat Assistant in the sidebar
- Chat naturally with Claude AI:
- "I'm a single mom with a 3.2 GPA and can afford $15k/year. Help me find colleges."
- "Why was this school recommended for me?"
- "What makes this college affordable?"
- "Find me schools within 50 miles of my home"
- Get personalized explanations and follow-up answers
- Navigate to 🗺️ School Map in the sidebar
- Toggle between:
- 🌎 All Schools - View all 6,800+ schools (blue circles)
- ⭐ My Recommendations - See your recommended schools (orange graduation cap pins)
- Filter by state using the dropdown
- Click any school to see detailed information:
- Institution name and location
- Public/Private type
- Admission rate (if available)
- Net price
- Median earnings (10 years post-enrollment)
- Your match score (for recommended schools)
- Navigate to 📊 Data Insights in the sidebar
- Explore three visualizations showing the income-graduation gap:
- Bar Chart - Average graduation rates by income level
- Scatter Plot - Individual schools showing correlation with trendline
- Box Plot - Distribution of outcomes by income category
- Understand why EquiPath was created
For each institution, we compute four composite scores:
1. ROI Score
roi_score = 0.6 * normalized_earnings + 0.4 * (1 - normalized_debt)- Uses median earnings 10 years after entry
- Uses median debt of completers
- Higher earnings and lower debt = better ROI
2. Affordability Score
# For standard students
afford_score_std = 0.6 * (1 - normalized_gap) + 0.4 * (1 - normalized_net_price)
# For student-parents (includes childcare costs)
afford_score_parent = 0.6 * (1 - normalized_gap_with_childcare) + 0.4 * (1 - normalized_net_price)- Lower net price and smaller affordability gap = more affordable
- Student parents face 3× larger affordability gap ($26,931 vs $9,610)
3. Equity Score
# Equity parity measures how equal grad rates are across races
equity_parity = 1 - (max_grad_rate - min_grad_rate) / 100
# Personalized equity score uses race-specific graduation rate
equity_score = 0.7 * race_specific_grad_rate_norm + 0.3 * equity_parity- If user is Black → uses Black graduation rate
- If user is Latino → uses Latino graduation rate
- Parity = 1.0: perfect equity; Parity = 0.0: maximum disparity
4. Access Score
access_score = normalized_admit_rate * fit_multiplier- Based on admission rate and GPA fit
- Safety schools (admit rate >70%) get boost for qualified students
- Reach schools (admit rate <30%) get penalty unless high GPA
# Base weights
alpha (ROI) = 0.25
beta (Affordability) = 0.30
gamma (Equity) = 0.25
delta (Access) = 0.20
# Adjustments based on user profile
if income == "LOW" or budget < 20000:
beta += 0.15 # Affordability matters more
alpha -= 0.05 # ROI matters less
if is_parent:
beta += 0.10 # Affordability critical
gamma += 0.05 # Equity support needed
if first_gen or race in ["BLACK", "HISPANIC", "NATIVE"]:
gamma += 0.10 # Equity support important
# Normalize weights to sum to 1.0
weights = normalize(weights)
# Final personalized score
user_score = alpha*ROI + beta*Affordability + gamma*Equity + delta*AccessBefore scoring, colleges are filtered by:
- Budget constraint: Net price ≤ 1.5× user budget
- Public only: If user specified
- In-state only: If user specified and state provided
- School size: If user has preference (Small/Medium/Large)
- Geographic radius: If user provided zip code and radius (e.g., 50 miles)
We use K-means clustering (k=5) with StandardScaler normalization on four features:
- ROI score
- Affordability score (standard)
- Equity parity
- Access score
Clusters are labeled based on centroid characteristics:
- High ROI + High Equity + Affordable → "Equity Champions"
- Affordable + High Access → "Accessible & Affordable"
- High Equity + High Access → "Equity-Focused Access"
- Balanced profile → "Balanced Options" (with numbered variants)
Random state is set to 42 for reproducibility.
UserProfile(
gpa=3.2,
budget=15000,
state="CA",
zip_code="90210",
radius_miles=50,
race="BLACK",
income_bracket="LOW",
first_gen=True,
is_parent=True,
in_state_only=True,
public_only=False
)Personalized Weights:
- Affordability: 55% (base 30% + 15% low income + 10% parent)
- Equity: 30% (base 25% + 5% parent)
- ROI: 10% (normalized after adjustments)
- Access: 5% (normalized)
Filters Applied:
- Net price ≤ $22,500 (1.5× budget)
- California schools only
- Within 50 miles of zip code 90210
Focus: Affordable institutions with strong support for parents and equity outcomes for Black students
UserProfile(
gpa=3.6,
budget=25000,
state="TX",
race="HISPANIC",
income_bracket="MEDIUM",
first_gen=True,
is_parent=False
)Personalized Weights:
- Equity: 35% (base 25% + 10% first-gen/Hispanic)
- Affordability: 30% (base, no adjustments)
- ROI: 22% (normalized)
- Access: 13% (normalized)
Focus: Institutions with strong equity outcomes for Latino students and reasonable costs
UserProfile(
gpa=3.9,
budget=50000,
race="ASIAN",
income_bracket="HIGH",
first_gen=False,
is_parent=False
)Personalized Weights:
- All weights remain at base (no adjustments)
- ROI: 25%
- Affordability: 30%
- Equity: 25%
- Access: 20%
Focus: Balanced search across all dimensions
python src/feature_engineering.py- Processes raw data and computes all scores
- Creates Parquet cache for fast loading
- Displays summary statistics
python src/clustering.py- Performs K-means clustering
- Shows cluster centroids and labels
- Displays sample institutions by cluster
python src/scoring.py- Tests personalized scoring algorithm
- Uses example profiles (low_income_parent, first_gen_student, etc.)
- Displays top 10 recommendations
# Requires ANTHROPIC_API_KEY in .env
python src/llm_integration.py- Tests Claude AI integration
- Generates explanations for recommendations
- 4,933 unique institutions in final dataset
- 6,812 schools on interactive map
- 288 features per institution
- 71.9% GeoJSON-to-database match rate
- Income-Graduation Gap: Schools with 75-100% Pell students graduate at significantly lower rates than schools with 0-25% Pell students
- Racial Equity Gaps:
- White students: 56.5% graduation rate
- Black students: 44.1% (12.4% gap)
- Latino students: 49.9% (6.6% gap)
- Asian students: 61.3%
- Student Parent Challenge: $26,931 median affordability gap (3× larger than standard students' $9,610)
- Completion Timeline: Only 40% graduate in 4 years; 53.1% in 6 years
- Highly Selective (<25% admission): 68 schools (1.4%)
- Moderately Selective (25-50%): 155 schools (3.1%)
- Open Access (>75%): 975 schools (19.8%)
- Average Admission Rate: 73.6%
- Equity First: Center the needs of historically underserved students
- Transparency: All scoring logic is auditable and explainable
- Data-Driven: Use official federal data, not opinions or rankings
- Personalization: Every student gets unique recommendations based on their profile
- Actionable: Provide clear, practical information for decision-making
- Honest: Show tradeoffs and real outcomes, not just "best matches"
- Accessible: Multiple interfaces (form, chat, map) to meet different user preferences
- Add more data sources to increase admission rate coverage (currently 35.4%)
- Include transfer student outcomes and pathways
- Add campus support services data (mental health, childcare availability, etc.)
- Integrate financial aid generosity metrics
- Train supervised ML model (XGBoost, Random Forest) to predict student success probability
- Add time-to-degree predictions
- Include intended major/career pathway recommendations
- Implement collaborative filtering (students like you chose...)
- Net price calculator integration
- Application deadline tracking and reminders
- Scholarship matching based on profile
- Peer comparison (outcomes for students like you at this school)
- Save/export recommendations as PDF
- Email integration
- Mobile app version
- User accounts to save search history and profiles
- Comparison tool (side-by-side school comparison)
- Virtual campus tours integration
- Student testimonials from similar backgrounds
This project was built for the 7th Annual Datathon - Track 2: Educational Equity.
Team Members:
- Alex Toohey
- Andre Lee
- Davyn Paringkoan
- Leo Zhang
This project uses publicly available federal education data from the U.S. Department of Education. The code and methodology are provided for educational and non-commercial use.
- Data Sources: U.S. Department of Education (College Scorecard, IPEDS), Affordability Gap dataset
- AI Partner: Anthropic (Claude API)
- Inspiration: The millions of students navigating higher education with limited resources and systemic barriers
- Mission: Making college advising more equitable, accessible, and effective for all students, especially those who need it most
- TECHNICAL_DOCUMENTATION.md - Comprehensive technical documentation with algorithms, data flow, and implementation details
- tableau_exports/TABLEAU_GUIDE.md - Guide for recreating visualizations in Tableau
Built with ❤️ for educational equity
"Education is the most powerful weapon which you can use to change the world." - Nelson Mandela