Inspiration
What it does
How we built it
Challenges we ran intoI’ve reformatted your exact text into a clean, professional Markdown structure. I have removed the references to the political initiative as requested while keeping your personal narrative and technical details intact.
VitalPath: Personalizing Health Education Health education became personal to me the moment I realized how little it actually adapts to the students it’s meant to serve. In U.S. schools, every student is taught the same health curriculum; regardless of their age, athletic involvement, cultural diet, injury history, or even the kind of food they have access to at home. I didn’t understand how much of a problem this was until I started talking directly to the people in my own school who work with students’ health every day.
Insights from the Front Lines Mrs. Purchatzke (Health Teacher): Shared how difficult it is to personalize lessons without insight into students’ real eating habits, growth patterns, or personal opinions.
Mr. Krupa (PE Teacher): Explained the struggle of accommodating different fitness levels and social pressures while ensuring students stay healthy without being bullied.
Ms. Rous (Athletic Trainer): Noted that many active students don’t understand the value of rest, often pushing through pain until they end up injured or burned out.
These conversations are why I built VitalPath.
What is VitalPath? VitalPath is my attempt to make health education feel relevant to each student, while respecting privacy and supporting teachers instead of overwhelming them.
For Students: Guidance that reflects their actual lives—what they eat, how they move, and how they recover.
For Educators: Meaningful insights without being turned into data analysts or disciplinarians.
For Schools: The ability to align fitness standards with real, anonymized trends instead of assumptions.
Technical Implementation To make this possible, I built a real AI system, not just a demo.
The Vision Pipeline When a student uploads a photo of their food, the app uses a multi-stage approach:
Detection: YOLOv8 to detect individual food items.
Classification: A fine-tuned EfficientNet-B3 model trained on Food-101.
Data Retrieval: Nutrition data is pulled from the USDA FoodData Central API.
Estimation: Portion size is estimated using custom heuristics based on bounding box area.
The UPF Exposure Score™ Instead of pushing calorie counting, which can be harmful for minors, I created a system to help students understand food quality. Using a hybrid rule-based and machine learning classifier grounded in NOVA food categories, each food is evaluated based on its processing level and additives. The goal isn’t to shame, but to help students reflect on weekly patterns.
AI Health Chatbot The platform includes a chatbot powered by OpenAI GPT-4o / GPT-5.2 with strict guardrails:
Uses Retrieval-Augmented Generation (RAG) with school-approved health sources.
Employs age-appropriate language and avoids medical diagnoses.
Integrates student data only with explicit consent to support habit-building.
Growth and Recovery Using FitnessGram data, VitalPath visualizes fitness trends over time with polynomial regression and time-series smoothing. Training load and rest gaps are analyzed using injury-risk heuristics, alerting teachers only when specific thresholds are crossed to ensure privacy.
Development Stack Frontend: React, TypeScript, and Tailwind CSS; Chart.js and Recharts for visualizations.
Backend: Node.js and Express (API Gateway); Python with FastAPI for AI inference.
Database & Auth: Firebase Authentication, Firestore, and AWS S3 for image handling.
Challenges and Accuracy One of my biggest challenges was accuracy. Early versions struggled with visually similar items, like coffee versus protein shakes. By splitting detection and classification into separate stages, I achieved:
93% identification accuracy in real-world testing.
82–88% top-1 accuracy on food classification test images.
12–18% average error on nutrition estimates, which is appropriate for educational use.
Conclusion: Responsible AI What makes VitalPath innovative isn’t just the technology—it's the intent. Despite obesity prevention being a multi-billion-dollar industry, few tools focus on personalization and education simultaneously.
Building this project taught me that AI in education must be transparent, non-judgmental, and privacy-first. VitalPath is my attempt to use AI the right way: to solve a real problem in my community by augmenting human judgment rather than replacing it.
Log in or sign up for Devpost to join the conversation.