Inspiration
My friends and I spend almost 15 hours a day on campus, and we hate missing out on free food events. We wished there was an easier way to get a list of events offering free food without scouring through email lists and Instagram accounts.
What it Does
The initial idea was to extract data from multiple sources (Instagram, Discord, Emails), but due to scope limitations, we’ve focused on Instagram for now. The web app is designed to provide a daily, refreshed list of events happening around campus, specifically those that offer free food.
How We Built It
I used Apify's Instagram scraper to gather account data, recent posts, captions, and URLs. Each post's URL was then used to download a local copy and run Google Vision's OCR to extract any relevant text from the image. Finally, I used Google Gemini's API to analyze the extracted data from both the captions and OCR results. This allowed us to classify events based on attributes like timing, location, and whether free food is available, which is then stored in a text file to be displayed on the web app.
Challenges We Ran Into
Planning out this project was challenging due to the different tools and APIs involved. One major obstacle was hitting the limits of free-tier APIs, which often slowed progress. Additionally, refining the AI responses was tricky; even after careful data engineering, the AI sometimes gave inconsistent outputs, requiring more specific prompts.
Accomplishments We're Proud Of
As a solo developer working on my first-ever hackathon and personal project, I’m proud of persevering through challenges, especially when things seemed to fall apart at each stage. I’m happy with the idea, the data processing pipeline, and the APIs I was able to set up, even with the limited resources.
What We Learned
This project taught me how quickly public Instagram data can be scraped and used to make inferences, which was both fascinating and concerning. I also learned the limitations of relying on AI for classification tasks—while it helped, the responses were often inconsistent, and the time spent fine-tuning prompts was significant.
What's Next for FindMyDinner
Originally a proof of concept, FindMyDinner has the potential to scale to include data from over 500 student clubs across the university. By actively scraping social media accounts, the project will improve student engagement, help them find events with free food, and serve as a great tool for networking and meeting new people.
Log in or sign up for Devpost to join the conversation.