Inspiration

People often can't read the small text on prescription bottles. This causes medication mistakes and confusion. We wanted to create an app that instantly reads prescription labels using AI so everyone can understand their medications safely.

What it does

The app lets you take a photo of a prescription bottle and automatically reads all the text on it. It shows you what it found and how confident it is about each word. Soon it will also research your medication, create simple explanations, and find nearby pharmacists for questions.

How we built it

We created a website with React (frontend) and Flask (backend). We used EasyOCR technology to read text from images, even curved text on bottles. We added image processing to reduce glare and improve accuracy. Everything runs on your computer with no subscription needed.

Challenges we ran into

Getting OCR to work on Windows took multiple tries with different libraries. Curved text on bottles is hard to read. Making the app handle blurry photos and bad lighting was tricky. Converting data to JSON format caused errors. But we solved each problem one by one.

Accomplishments that we're proud of

We built a complete, working app from scratch. The code is clean and well-documented. The design looks professional and is easy to use. Everything handles errors gracefully. We created a plan for advanced features like AI research and specialist finder.

What we learned

Different OCR tools work better for different situations. Image improvement (denoising) helps more than fancy AI models. Real prescription photos are messy and hard - not like test images. Good documentation and error messages matter as much as the code itself.

What's next for RxCompass

We'll add automatic research of side effects using PubMed. Then we'll create AI summaries doctors can actually read. After that, we'll find local pharmacists and doctors nearby. Eventually we want mobile apps and user accounts to save prescription history. The goal is to make medication information accessible to everyone.