-
-
GravityShield – AI Early Risk Intelligence Platform
GravityShield – AI Early Risk Intelligence Platform
🌍 Inspiration
Across tier-2 and tier-3 institutions, many students silently struggle with declining academic performance, emotional stress, and financial pressures. Most institutions react only after visible failure — when dropouts or severe mental health issues have already occurred.
We were inspired by a simple idea:
What if we could detect the fall before it happens?
Just like gravity pulls objects downward, academic and emotional risks pull students toward crisis. GravityShield acts as an early-warning AI system that detects risk trends and lifts students upward through timely, personalized intervention.
🚀 What It Does
GravityShield is an AI-powered Early Risk Intelligence Platform that:
- Predicts academic risk using attendance, marks, engagement, and financial indicators
- Analyzes student feedback using NLP-based sentiment detection
- Generates a combined risk score ( R \in [0,100] )
- Classifies students into Low, Medium, or High risk categories
- Automatically generates personalized intervention plans
- Provides mentor alerts for high-risk cases
- Displays institutional risk trends via an interactive dashboard
Instead of reacting to failure, GravityShield enables proactive prevention.
🧠 AI Risk Engine
1️⃣ Academic Risk Model
We trained a Random Forest classifier using:
- Attendance ( A )
- Internal Marks ( M )
- Assignment Delay ( D_a )
- Fee Delay ( D_f )
- Engagement Score ( E )
The model outputs an academic probability:
$$ P_{academic} \in [0,1] $$
2️⃣ Sentiment Intelligence
Using NLP-based sentiment scoring:
$$ S_{emotion} \in [-1,1] $$
We normalize emotional polarity into risk probability:
$$ P_{emotion} = \frac{1 - S_{emotion}}{2} $$
3️⃣ Unified Risk Score
We combine structured and unstructured intelligence:
$$ R_{final} = \left( w_1 \cdot P_{academic} + w_2 \cdot P_{emotion} \right) \times 100 $$
Where:
- ( w_1 + w_2 = 1 )
- ( w_1 ) = Academic weight
- ( w_2 ) = Emotional weight
4️⃣ Risk Classification
$$ \text{Low Risk} \quad (0 \leq R < 40) $$
$$ \text{Medium Risk} \quad (40 \leq R < 70) $$
$$ \text{High Risk} \quad (70 \leq R \leq 100) $$
The engine also outputs:
- Contributing risk factors
- AI-generated personalized intervention roadmap
🛠 How We Built It
Backend
- FastAPI for API development
- Scikit-learn Random Forest model for academic prediction
- TextBlob NLP for sentiment analysis
- SQLite for lightweight database storage
We generated a synthetic dataset of ( n = 300 ) student records to train our classification model.
Frontend
- React + Vite
- Tailwind CSS for clean UI
- Framer Motion for Anti-Gravity themed animations
- Chart.js for institutional risk visualization
The system is designed to scale and deploy on platforms such as Render and Vercel.
⚠ Challenges We Ran Into
- Creating realistic synthetic datasets for ML training
- Maintaining consistent feature ordering between training and inference
- Handling neutral sentiment inputs ( S_{emotion} \approx 0 )
- Managing CORS between frontend and backend
- Balancing AI sophistication with hackathon feasibility
- Designing measurable social impact metrics
Each challenge strengthened technical robustness and real-world applicability.
🏆 Accomplishments
- Built a working AI-powered early risk prediction model
- Integrated structured + unstructured intelligence
- Developed a unified probabilistic risk engine
- Designed measurable KPIs for social impact
- Created a scalable institutional prototype
- Ensured ethical AI usage with explainable outputs
Most importantly, we moved from idea to a functional, demo-ready system.
📚 What We Learned
- AI is most powerful when used for prevention, not reaction
- Combining structured and unstructured data improves predictive strength
- Social impact solutions require measurable outputs
- Simplicity improves adoption
- Responsible AI must include transparency and explainability
Innovation happens at the intersection of technology and empathy.
🔮 What's Next
- Pilot implementation in partner colleges
- Expand dataset with real institutional data
- Improve accuracy using ensemble learning
- Add parent & mentor notification modules
- Integrate scholarship and career recommendation systems
- Deploy multilingual support
- Explore government and NGO partnerships
Long-Term Vision
To develop GravityShield into a national-level AI guardian system for student well-being.
Log in or sign up for Devpost to join the conversation.