Hack-A-Food

upload your food!!!
food dashboard!
snacks, snacks, snacks!!!

Inspiration 🤒

Hackathons are objectively one of the most unhealthy activities for humans 🍕🍟. When Tom got a stomach bug 🤢 after the first night (not Hacklytics' fault 😅), we wanted to investigate what was really in our food. We believe food is the most dangerous drug in America 💊🇺🇸, and by giving people an easy, firsthand view 👀 of their consumption, everyone can make better, more sustainable 🌱 munching options.

What it does 🔍

Hack-A-Food essentially "hacks" into any packaged food identified by a barcode 📦📱. Upon scanning the package, the user gets insights on all ingredients 🥦🥩, nutrients 💊, additives ⚗️, as well as environmental 🌍 influences involved in creating the food. Our website also allows for interactive learning through Cerebras Inference 🧠 and information marks. We even recommend similar foods 🍽️, enabling users to explore and discover healthier or more interesting alternatives as they browse through the web app.

How we built it 🛠️

We utilized the Open Food Facts dataset, providing comprehensive attributes for over 3.7 million varieties of packaged food. We called their API 🌐 to access information based on food barcodes. For our food similarity search algorithm, we employed Term Frequency-Inverse Document Frequency (TF-IDF) 📊 vectorization of product names and keywords, calculating cosine similarity to recommend similar products. This approach effectively leveraged names, keywords, nutri-score 📈, and food attributes, with adjustable weights ⚖️. Our frontend was built with the Next.js framework 💻 for a reactive, responsive web experience.

Challenges we ran into 😅

Deployment. After finishing, we spent about ~4 hours struggling with deployment, yet still couldn't fully succeed 😵‍💫. Technical debt 🧾 within our system led to infinite loops of errors, despite successful builds. The web interface itself deployed partially, but the backend infrastructure hit issues like import/export errors ⚠️. We also aimed to create a reinforcement learning (RL) model to understand relationships between foods, but computational constraints with 3.7 million data points made it impractical 🤔. We attempted a smaller-scale version and achieved decent results. Ultimately, we settled for a simpler recommender system but enjoyed experimenting with RL.

Accomplishments that we're proud of 💪

We completed a hackathon despite getting super sick 🤧—a true test of perseverance. We're proud of experimenting with reinforcement learning and successfully developing a smaller-scale (~10k data points) recommender model. It was incredibly rewarding to explore these new concepts.

What we learned 🧠

It's never truly over until it's over 🕒. Even when feeling down 😔, pressed for time, and unsure we'd finish, we quickly adapted to challenges through resourceful teamwork and technology ⚡. Resilience and flexibility were key lessons 🔑 we took from this hackathon.

What's next for Hack-A-Food 📱

Mobile App 📲! We initially attempted mobile development with Expo and React-Native but found the learning curve too steep 📈. Now, without hackathon constraints, we're determined to create an accessible, user-friendly mobile app 🌟. Promoting food health education 🍎📚 should be simple and accessible to everyone!