Celio - Shopping Revolutionized

🧠 Inspiration

Shopping hasn't really changed in the last 20 years. Sure, online stores got slicker and you can now tap your phone to pay, but at its core? Still the same chaos. Stats say that 70% of shoppers research online before buying, and over 80% of Gen Z shoppers say they feel overwhelmed when trying to pick the “right” item. And for people with social anxiety? Asking for help in-store is not ideal (ahem mcdonalds uses kiosks for a reason!). Instead, they open 30+ tabs, compare Reddit posts, TikTok reviews, price trackers, and maybe—just maybe—check out. We thought: there has to be a better way. So we built it.

🤖 What it does

Meet Celio — your AI-powered shopping assistant that lives in the real world. Think Costco, IKEA, or any store where you wander around like you're in a maze. Celio is built for enterprise, companies can install it in physical stores to guide customers, answer questions, give product recommendations, and even personalize based on your vibe. It's like ChatGPT with a shopping PhD and a GPS.

🛠️ How we built it

Okay, you got us. At first glance it might look like another wrapper, but we promise, it's deeper than that. We built:

Frontend in React and Shadcn for a smooth, responsive chat UI
Backend with AWS Lambda to keep things light and serverless
EC2 instance to host the AI and WebSocket server — so responses are instant (no 10-second delays like what you see with other AI wrappers!)
A real-time WebSocket connection to keep the convo flowing without reloads or lag
Word2Vec-style embedding models to capture sentiment and intent behind what users type
k-NN (k-nearest neighbors) algorithm with OpenSearch, a vector DB, to find the best product matches, fast

The result? A smart, responsive shopping assistant that doesn’t just answer questions, it gets you (wink).

🧩 Challenges we ran into

EC2 setup was a PAIN. Between firewalls, IAM roles, and managing open ports on the system, we spent hours figuring out Amazon’s idea of “secure” (and trying to bypass it).
Making AI respond in real time required some serious tweaking, especially getting the WebSocket architecture working reliably.
And of course: prompt engineering. Getting our model to give consistent, helpful replies meant crafting 300+ word prompts, adjusting, testing, and praying (english class was useful after all).

🏆 Accomplishments we’re proud of

We went way beyond just calling an AI API (we are looking at you Cluely..). We built a full backend pipeline; from user input ➡ embedding ➡ vector search ➡ product output, all wrapped in a sleek UI. A lot of hackathon projects focus on the frontend "wow" factor. We focused on scalability, real-world infrastructure, and actual product viability, the very emphasize of this hackathon.

📚 What we learned

Sleep = debugging superpower, y'all should genuinely hop on
Caffeine = 10x dev mode
Don’t show up early for food. Seriously, wait 2:15 minutes and skip the (shawarma) line. Also: vector databases are awesome. WebSockets are weirdly satisfying. And making AI feel human is hard, but so worth it.

🚀 What’s next for Celio — Shopping Revolutionized

We want to expand beyond shopping. The same system could work in museums, airports, hospitals, or anywhere people need fast, friendly info without waiting in line. We're also thinking multilingual support, voice interaction, and maybe a mobile AR version. The way we wrote the code makes it super robust, modular, and future-proof.