SmartStation

Inspiration

SmartStation was inspired by real-world airline catering and logistics pain points, such as manual inventory checks, frequent scanning errors, and a lack of accessible supervisor analytics. We sought to build a lightweight, unified system to solve these problems, combining a vision-based analyzer (that instantly verifies alcoholic beverages against airline regulations) with AI-assisted visual checks and real-time productivity metrics. The goal was to reduce errors, expedite turnarounds, and optimize the managers' work. In other words, we were inspired by the opportunity to build an all-around solution that not only automated the assembly line process through inventory management but also provided real-time feedback for supervisors, all managed through a connected database and mobile app. This was a project like we had never done before, and we believe it has helped us grow both professionally and personally.

What it does

SmartStation automates flight inventory validation and operator workflows on a single unified website. Employees can enter the flight number, then automatically get the inventory needed for that flight divided into easy-to-package carts, after packaging, they only need to take a photo of each drawer of the cart and the system automatically scan items (images + weight), then the system uses vision + heuristics to validate expected inventory, and the backend returns clear pass/fail results. Also, it automates the task of alcoholic bottle handling, using a simple photo of the bottle and putting the airline, it automatically tells what to do with that bottle based on the different policies that each airline has, helping the employees with the decision-making process, it is most needed. Supervisors get an analytics dashboard with productivity (drawers/hour), error rates, and actionable insights to balance workloads and prioritize training, all this while integrating an AI assistant for personalized and optimal solutions or analysis, saving time and making the workflow more efficient for the supervisor.

How we built it

Frontend: React + Vite + Tailwind for a fast, responsive UI and role-based routing. Backend: Node.js + Express with JWT auth, secure HTTP-only cookies, and role middleware. Database: Snowflake for production analytics and persistence (init script can populate demo data). Scripts: Python utilities to seed inventory and validate or simulate scans. AI: Image detection integration hooks, integrating the Gemini api to extract detected product types and measured weight from photos. DevOps: Environment variables for secrets, CORS configured, and simple init scripts for demo data.

Challenges we ran into

Simulating reliable vision-based detections for a demo environment — solved with mock outputs and a clearly defined JSON contract. Secure token handling across frontend/backend while keeping the demo easy to run locally (balanced HTTP-only cookies and CORS). Designing supervisor metrics that are informative but not noisy — required iterating on aggregation queries and thresholds. Making the inventory UX fast and forgiving for operators (tolerances, re-scan flow). Accomplishments that we're proud of End-to-end flow: secure login → role-aware UI → inventory scanning → backend validation → supervisor analytics. Clear, production-aware security choices (bcrypt hashing, HTTP-only JWT cookies, parameterized queries). A compact init script that seeds demo users and metrics so stakeholders can test features locally in minutes. Practical dashboard features (drawers/hour, error rates, automated performance badges) that deliver actionable insights.

What we learned

In this project, we put our knowledge of UX and UI interfaces to the test; however, we were surprised to learn that small UX improvements could be made, thanks to the guidance of an AI agent. Therefore, small, focused UX improvements (tolerance messaging, category checks) were implemented, dramatically reducing operator mistakes. We also learned how to work on a centralized workflow, meaning that one person was in charge of one part of the webpage.

We were also able to practice teamwork and see firsthand how it really does make the dream work. Since we each had our own assigned tasks, we couldn't just put our heads down and work alone. We learned very quickly that clear communication and a strong sense of reliability weren't optional—they were crucial. We constantly had to know what others were doing and trust that everyone was handling their piece of the puzzle. This forced us to get really comfortable with the "done is better than perfect" mentality. Learning to be accountable and deliver on time, even if you're a perfectionist and know it's not your absolute best work, is what builds trust. That trust was the glue that held our team together.

While researching, we found out supervisors prefer clear, prioritized recommendations instead of raw numbers; invest time in insight design, not only charts. We understood that supervisors are busy and don't have time to be data analysts. A dashboard full of line graphs and pie charts just adds cognitive load. They don’t want to interpret the data; they want the system to do it for them. We found it’s better to invest time translating raw outputs into plain English, prioritizing recommendations. Instead of showing 'Error Rate: 15%,' it's infinitely more valuable to show 'Top Action: Retrain Station 3 on beverage placement.' This is an investment in insight design, not just data visualization.

Before this hackathon, our team's strength lay deeply in frontend and some backend development, but we weren't off to a good start with the database implementation. We struggled with the initial setup and weren't sure how to best structure our data. However, after a few tries, we decided to pivot and truly commit to implementing Snowflake's tools and power as best we possibly could. This decision changed everything. We learned a massive amount about databases, not from textbooks, but through prompt engineering.

What's next for SmartStation

Further train the gemini api to have a higher confidence value. Add audit logging, user management, and admin tools for production rollouts. Implement refresh tokens and optional multi-factor authentication. Add exportable reports (PDF/CSV), real-time sync for offline-first operations, and mobile-first scanner UI. Improve metrics with trend forecasting and anomaly detection to identify issues before they affect operations.