PCOSense

🧠 Inspiration

Polycystic Ovary Syndrome (PCOS) affects 1 in 5 women in India and remains widely misunderstood. We wanted to build a tool that not only helps manage the symptoms but empowers women with AI-driven insights, personalized care, and a sense of support. Our goal was to create an app that bridges the gap between complex medical data and everyday self-care, especially in a society where such conversations are often silenced.

💡 What it does

PCOSense is a smart health companion for women with PCOS that provides:

• AI-powered cycle prediction and symptom tracking.
• AI-based food scanner that assesses both cooked and packaged food for PCOS-friendliness using a knowledge graph.
• Yoga and mindfulness sessions with real-time AI pose matching.
• Backend intelligence using Neo4j, MongoDB, ChromaDB, and LLM inference for personalized care suggestions.

🛠️ How we built it

We used a Flutter frontend and integrated it with:

• Firebase Authentication for user management.
• FastAPI backend with Python for inference, data handling, and REST APIs.
• MongoDB Atlas to store user images, logs, and nutritional records.
• Neo4j to model dietary relationships and support recommendation queries.
• ChromaDB to store vector embeddings and perform semantic search on food descriptions and user queries.
• Google ML Kit and custom ML models for pose detection, OCR, and food analysis.
• Gemini API and Ollama for large language model (LLM)-based inference.
• AI tools like ScraperAPI and Hugging Face for structured web scraping and language tasks.

🧗 Challenges we ran into

• Getting high-quality, structured food data was a major hurdle — we had to scrape data across multiple websites and consolidate inconsistent nutrition and ingredient formats.
• Designing a knowledge graph and embedding system that supports contextual food recommendations.

🏆 Accomplishments that we're proud of

• Successfully integrated cooked food and packed food analysis using AI.
• Created a custom recommendation engine with Neo4j and ChromaDB vector search.
• Developed a complete end-to-end mobile app with backend and AI workflows .
• Created a solution that can genuinely improve lives through personalization and empathy.

📚 What we learned

• How to integrate multiple AI pipelines (food OCR, pose detection, vector search, recommendation) into a mobile workflow.
• Hands-on experience with Neo4j, ChromaDB, Google ML Kit, Gemini API, and Ollama for inference.
• Importance of user-centric design when solving sensitive healthcare problems.
• How to work under pressure as a cross-functional team in a real hackathon environment.

🔮 What's next for PCOSense

• Enhance the recommendation engine using reinforcement learning and ChromaDB semantic expansions.
• Incorporate telemedicine APIs for easy doctor consultations.
• Integrate with wearables for real-time data from fitness trackers.

GitHub Repository: https://github.com/Jothika1526/PCOSense

Built With

• fastapi
• firebase
• flutter
• mongodb
• neo4j
• ollama