SmellSense AI

• 

Inspiration

The inspiration for SmellSense AI came from a single, powerful question: What if we could detect life-threatening diseases earlier, using a method as simple and non-invasive as breathing? We were fascinated by the real-world science of volatolomics, where researchers analyze the Volatile Organic Compounds (VOCs) in human breath to find chemical signatures of diseases like cancer and diabetes. The high human and financial cost of late-stage diagnosis is a critical global problem, and we were inspired to explore how AI could make this cutting-edge science more accessible. Our goal was to build a tool that not only demonstrates this potential but also educates users on the science behind it.

What it does

SmellSense AI is a fully functional, web-based proof-of-concept that simulates how Artificial Intelligence can perform early-stage disease screening. Our application features an interactive "AI Test Bench" where users can:

Load simulated "breathprint" data from different patient profiles: Healthy, Diabetic, and Lung Cancer.

Initiate an AI scan of the chemical biomarkers (VOCs) in the sample.

Receive a real-time diagnosis from the AI model.

View a dynamic graph that visualizes the unique chemical signature for each condition, making the AI's decision process transparent and understandable.

In essence, SmellSense AI translates a complex scientific concept into a tangible and interactive experience, showcasing the future of non-invasive diagnostics.

How we built it

We built SmellSense AI as a full-stack web application in an intensive 8-hour sprint.

Backend: The server is powered by Python using the Flask framework. It exposes a simple REST API endpoint (/predict) that receives breath sample data in JSON format.

Machine Learning: We used Scikit-learn and Pandas to train a RandomForestClassifier model. We created a simulated dataset in a CSV file, where each row represents a breath sample with varying concentrations of 10 key VOCs, to train the model to accurately differentiate between the three health profiles.

Frontend: The user interface is built with standard HTML, CSS, and vanilla JavaScript. To create a modern and responsive experience, we handled all API communication asynchronously using the fetch() API. The dynamic and interactive graph that visualizes the breathprints is rendered using Chart.js.

Challenges we ran into

Our biggest challenge was not technical but conceptual: credibility. How could we convincingly demonstrate a medical AI without a physical sensor or real patient data? A simple "fake" demo would feel dishonest and uninspiring. We solved this by pivoting from a simple UI to the "AI Test Bench" concept. This reframed the project as an honest and transparent simulation, allowing us to focus on the AI's analytical power rather than pretending we had a finished medical device. This approach turned our biggest weakness into a compelling narrative strength.

Accomplishments that we're proud of

From Idea to Full-Stack in One Day: We are incredibly proud of architecting, building, and integrating a complete application—backend, ML model, and frontend—in a single work session.

The "AI Test Bench" Concept: We feel this was our most significant accomplishment. It's an honest, credible, and engaging way to demonstrate a complex AI system, making the underlying science accessible to everyone.

Scientific Grounding: We successfully connected every feature of our prototype back to real-world scientific principles, creating a project that is both technologically impressive and scientifically sound.

A Polished, Accessible UI: We implemented a clean, modern user interface with key accessibility features, ensuring the demo is usable and understandable for a wide audience.

What we learned

Throughout this hackathon, we learned that a project's narrative is as important as its code. Grounding our technology in a compelling, real-world scientific story made the final product exponentially more impactful. We also reinforced the value of rapid, iterative design; our UI and demo concept improved dramatically with each iteration. Finally, this project was a masterclass in focusing on a core, achievable goal and executing it flawlessly under a tight deadline.

What's next for SmellSense AI

The journey for SmellSense AI is just beginning. Our vision for the future includes:

Hardware Integration: The highest priority is to connect our AI with real, low-cost electronic nose (e-nose) sensors to analyze live breath samples and move beyond simulation.

Model Expansion: We plan to acquire larger, anonymized clinical datasets to train the model to detect a wider range of conditions, such as kidney disease, liver failure, and other forms of cancer.

Clinical Validation: In the long term, we aim to partner with medical research institutions to validate our model's accuracy against traditional diagnostic standards and work towards a clinically viable screening tool.

Built With

• es6
• flask
• interactivity
• javascript
• pandas
• python
• scikit-learn