BiteWise

Inspiration

Food labels are confusing, receipts are ignored, and health insights are often locked behind expensive tracking apps that require manual input or invasive bank access. We realized that receipts already contain a complete, honest record of what people actually eat; however, not many people think of actually using them.

BiteWise was inspired by the idea that health feedback should be more immediate and easily accessible to people. Instead of asking users to log meals or connect financial accounts, we wanted to transform something people already have, their grocery and restaurant receipts, into actionable and tangible health insights.

What it does

BiteWise turns food receipts into health insights.

Users upload or scan any grocery or restaurant receipt, and BiteWise:

Extracts purchased food items directly from the receipt
Flags allergens (e.g., dairy, peanuts, gluten), with high-risk warnings based on user preferences
Identifies nutrition risks like high sugar, high sodium, or highly processed foods
Generates a clear health summary and score (0–100) with explanations
Offers smart suggestions, such as healthier swaps and next-time tips
Graphs insights into a weekly dashboard with trends, alerts, and summaries

How we built it

We built BiteWise using Figma for design, a Next.js frontend, a FastAPI (Python) backend, MongoDB Atlas for data storage, and the Google Gemini API for AI-powered analysis.

Challenges we ran into

Some receipts have messy formats, so it is difficult to scan them and successfully output the correct format
Many receipts list shortened or unclear product names, making classification difficult

Accomplishments that we're proud of

Built an end-to-end receipt-to-health pipeline that works for both grocery and restaurant data
Made the website accessible for people who have trouble reading (with the audio feature) as well as a text enlargement feature
Designed a clear and organized website that prioritizes allergens and safety first
Created a health score with explanations, not just a black-box number

What we learned

Designing clear user flows and interfaces using Figma
Building a responsive frontend with Next.js, TypeScript, and Tailwind CSS
Structuring React components
Using FastAPI (Python) to build and organize backend APIs
Storing and querying semi-structured data with MongoDB Atlas
Integrating Google Gemini API to parse OCR output and generate health insights

What's next for BiteWise

Deeper nutrition analysis using branded product databases
Optional integrations with other types of apps (for example: wellness apps)
Meal-level pattern detection across multiple receipts

Built With

backend:-fastapi
gemini-api
mongodb-atlas.-python-frontend:-next.js
react
tailwind-css
typescript
web-speech-api

Submitted to

DeltaHacks 12

Created by

Building the frontend for BiteWise focused on creating a clean, modern, and highly usable experience. I designed and implemented the UI using a component-based approach, ensuring consistency across pages while keeping the layout intuitive and visually clean. A major goal was making complex data, like receipt breakdowns, health scores, and spending insights, easy to understand at a glance.

Accessibility was a core consideration from the start. I implemented full dark mode support and added screen-reader announcements for key actions such as editing items and completing analysis. All interactive elements are keyboard-navigable, and forms use clear labels and feedback to reduce user friction.

I also optimized the interface for mobile and small screens, carefully refining spacing, input sizing, and layouts so receipt editing and scanning workflows feel natural on any device. Overall, this project taught me how thoughtful UI/UX decisions, accessibility, and responsive design directly impact how approachable and trustworthy an application feels.

Jason Yang
Building the backend for BiteWise was my first experience working with FastAPI and MongoDB Atlas. It was both challenging and incredibly rewarding. I had to learn how to structure a production-ready API from scratch, implementing 19 different endpoints that handle everything from file uploads to complex health analysis.

Working with MongoDB for the first time taught me about database design and asynchronous operations using Motor. I created four collections to manage receipts, user profiles, authentication, and API keys, learning how to properly index and query document-based data. The authentication system was my first attempt at implementing password hashing and token-based access control.

The most exciting challenge was integrating Google's Gemini AI API for the first time. I had to learn prompt engineering to get accurate OCR results from receipt images, structuring my requests to return properly formatted JSON with merchant names, dates, items, and prices. When I discovered the API had rate limits and could return 503 errors, I implemented a multi-model fallback system and learned how to handle failures by falling back to local Tesseract OCR.

Building the OCR parser was particularly educational. I created pattern matching algorithms to handle five different receipt formats, from simple "item price" layouts to complex weight-based pricing. I implemented guardrails that validate and ensure prices are correct, catching OCR errors such as absurd quantities or negative prices.

Sneakybarv
Rehan Patel
Pritika Lahiri

Updates

Pritika Lahiri started this project — Jan 11, 2026 01:00 AM EST

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.