SnapChef
Where Every Meal Begins with a Snap.
The problem SnapChef solves
While building the application, we faced several technical challenges. One major issue was that the frontend and backend were not producing the expected output, which affected the flow of data and user interaction. The backend's image processing model also encountered errors while trying to identify ingredients from uploaded images, possibly due to compatibility or model limitations. Additionally, integrating various parts of the program—connecting the model, backend logic, and frontend UI—proved to be more complex than expected within the limited time, causing delays and miscommunication between components. These issues highlighted the challenges of working with real-time AI pipelines under tight deadlines.
Challenges we ran into
We initially attempted to train or fine-tune a machine learning model ourselves. However, during this process, we encountered significant challenges; primarily the time and computational resources required to train the model using our dataset. The training process proved to be inefficient and impractical for our current setup.
As a result, we explored alternative solutions and discovered a more efficient approach: leveraging the Gemini API to identify ingredients directly from images. This API offered a faster, more reliable method for image-to-text processing, allowing us to bypass the lengthy training phase. By integrating the Gemini API into our system, we could extract ingredient information from food images and seamlessly pass that data to our backend for further processing.
This approach not only improved performance and accuracy but also allowed us to focus on refining other aspects of the project rather than getting bogged down in training complexities.
Tracks Applied (1)