Inspiration

We’ve all been there—you’re scrolling through Instagram or TikTok and come across a video of a perfectly sizzling bowl of ramen or the crispiest chicken sandwich you’ve ever seen. You immediately send it to your group chat with, “WE NEED TO GO.” But then… nothing happens. Maybe it gets lost behind a flood of memes, or no one follows up, or the logistics of finding a time that works for everyone kills the vibe. Foodie inspiration is everywhere, but turning that excitement into actual plans is surprisingly hard. We built Triply to fix that. It’s your personal food adventure planner: automating everything from saving restaurant posts to booking the hangout and even splitting the bill afterward. What It Does Triply turns a social food recommendation into a real outing, effortlessly. Here’s how it works: Create a Party: A “party” is your friend group; think of it like your foodie crew or group chat. Drop a Post: Paste an Instagram post into the party. It could be a reel, a picture, or a carousel. AI Analysis: Our system uses AI to analyze the content (via screenshot) and extract: Restaurant name Location Food items Tags, vibes, and other context clues Restaurant name Location Food items Tags, vibes, and other context clues Scheduling the Hangout: Based on the restaurant's open hours and all party members’ availability, Triply suggests dates and times for the visit. Bill Management: After the hangout, users can upload a photo of the receipt. Our system parses it, identifies who ordered what, and intuitively splits the bill. No more awkward math or Venmo chasing! Triply isn’t just a restaurant recommender. It’s a full-circle tool to turn spontaneous foodie hype into actual memories. How We Built It Frontend: Built with Next.js, designed to be minimal and mobile-first for seamless use while texting or browsing social media. AI Backend: We use Claude to process screenshots of Instagram posts. It extracts structured data from unstructured visual media using carefully tuned prompts. We incorporate Google Maps APIs for resolving locations, hours, and directions. We use Claude to process screenshots of Instagram posts. It extracts structured data from unstructured visual media using carefully tuned prompts. We incorporate Google Maps APIs for resolving locations, hours, and directions. Bill Splitting: We use Claude to process screenshots, filter out irrelevant entries, and tag items for each person using menu recognition and named entity detection. We use Claude to process screenshots, filter out irrelevant entries, and tag items for each person using menu recognition and named entity detection. Data Handling: User data is stored on Firebase and used only to optimize planning and payment flows. User data is stored on Firebase and used only to optimize planning and payment flows. Challenges We Ran Into Our biggest challenge was post analysis. Instagram is notoriously bot-resistant, so we had to rely on screenshot inputs instead of scraping text or metadata. Screenshots are messy (there’s no standard format) but we engineered a system to consistently pull relevant info with high reliability. Other major challenges: Balancing Claude’s token limits, input fidelity, and number of images. Too little detail, and the model fails; too much, and it takes overly long to process. Building prompts that generalize across a variety of post types (reels, stories, photos, captions). Accurately identifying locations, even for lowkey or niche restaurants, and handling near-duplicate or misspelled names. Creating a user-friendly and reliable receipt splitting tool that works without requiring every user to itemize their own order. Accomplishments We’re Proud Of Our AI post analysis works surprisingly well. Even on visually chaotic posts or vague captions, it often extracts the correct restaurant and food items. We built a working prototype that turns a post into a scheduled plan end-to-end. We tackled an actual real-world pain point that we experience and made something we’d genuinely use. We created an intuitive UI that doesn’t require users to understand the tech under the hood. What We Learned Prompt engineering is no joke. It can be make or break for a technology like this. You can spend hours tuning just a few words to make a model behave better. Planning things with friends is a surprisingly deep problem space. There’s so much room for intelligent tooling. This was our first hackathon! Git was interesting at times. What’s Next for Triply We’re excited to expand Triply’s capabilities: Menu Matching: Automatically cross-reference food items with real restaurant menus for better recommendations and receipt tagging. Multi-platform Support: Bring in content from TikTok, Yelp, YouTube, and Google Reviews. Auto-invite & Calendar Integration: Instantly create calendar events with reservation links and reminders. Group Polls & Preferences: Let groups vote on where to go next, with intelligent ranking based on past hangouts. Food Journal Mode: Turn your visits into a shared food diary, with ratings, photos, and comments. Triply is just getting started. We want to make group food adventures effortless and way more fun! Slides Check out our demo slides here: Triply Demo Presentation