Gate Keeper

Inspiration

One of gategroup's biggest strengths and vulnerabilities is its workforce. Every day, thousands of flight trolleys are packed manually across the world by their staff. Each trolley contains dozens of drawers, and each drawer must have specific products in exact quantities, positioned precisely according to airline specifications. One mistake such as a missing water bottle, a wrong snack in business class can cascade into passenger complaints, airline penalties, and operational chaos. The challenge? Human error is inevitable when operators pack 200+ drawers per shift under extreme time pressure with only paper stickers as guidance. We were inspired by the impossible position these operators are placed in: cognitive overload from remembering hundreds of product combinations, zero margin for error once a trolley is sealed, and no digital feedback system to catch mistakes before they reach the aircraft. We asked ourselves: what if every packing station had intelligent eyes? What if technology could see what operators pack in real-time, compare it against specifications instantly, and alert them before the drawer is closed?

What it does

Gate Keeper is a real-time error detection system that validates every drawer the moment it's packed. The web-based platform guides operators step-by-step through flight preparation: they select their assigned flight, choose a trolley, pick a side (A or B), and work through each drawer sequentially. For every drawer, the system displays the exact products and quantities required. Once the operator fills the drawer physically, they capture a top-down photo and enter the measured weight. Gate Keeper then uses computer vision AI to identify which items are actually present and compares both the detected contents and weight against the flight specification. If something's wrong, missing items, incorrect products, weight variance, the operator receives immediate feedback (OK, WARNING, or ALERT) with specific details like "Missing 2× Sparkling Water." They can fix the issue on the spot before moving forward. Every action is timestamped and linked to the operator ID, creating an auditable trail for compliance and performance analysis. The system replaces paper stickers with live, interactive validation that prevents errors from reaching the aircraft.

How we built it

We built Gate Keeper using a modern React + TypeScript stack with Tailwind CSS for a mobile-first, production-ready interface. The frontend is designed as a progressive web application optimized for tablets at packing stations, with large touch targets for operators wearing gloves, dark mode for reduced eye strain, and instant visual feedback through color-coded status badges. The data architecture mirrors gategroup's operational complexity: we designed a normalized Snowflake database schema that models flights, airlines, trolley specifications, drawers, products, and critically a Product_AirlineAvailability table that handles multi-airline product authorization without duplicating inventory. The validation workflow integrates computer vision (designed for Gemini AI API) to analyze drawer photos and compare detected items against expected configurations stored in the database. Weight verification provides a secondary validation layer. The entire system is built with RESTful API patterns in mind, making it ready to connect to a FastAPI backend that would handle AI inference, database queries, and event logging for real-time operational telemetry.

Challenges we ran into

Hours into the hackathon, gategroup released an updated brief that outlined their expected solution and it matched almost everything we had designed. Our "innovative" architecture was suddenly the baseline. We faced a critical decision: start over with a different angle, or prove we could actually execute what everyone else would only present as slides. We chose execution. With limited time remaining, we pivoted from comprehensive planning to surgical implementation. The biggest technical challenge was converting a complex operational workflow into an intuitive mobile interface that operators could use under pressure. We had to balance information density with simplicity showing enough detail without overwhelming users who need to work fast. Designing the database schema for multi-airline operations was another constraint: how do you model airline-specific product rules, flight specifications, drawer configurations, and validation events in a way that scales globally? We solved this through careful normalization and relationship modeling, ensuring data integrity while maintaining query performance. The final challenge was time: building a production-quality React application with proper state management, type safety, and responsive design in just hours required ruthless prioritization and clean architecture decisions.

Accomplishments that we're proud of

We're proud that we built a working, production-ready prototype instead of a concept deck. While others presented ideas, we demonstrated a functional application with real navigation flows, validation workflows, and a user experience designed for the actual constraints of airline catering operations. The database schema we designed is robust enough to handle gategroup's real-world complexity multi-airline operations, product authorization rules, drawer-level validation tracking, and operator performance metrics. We're especially proud of the UI/UX decisions: dark mode for long shifts, clear status indicators that work at a glance, and a step-by-step workflow that reduces cognitive load. The system respects the operator's reality. they're not sitting at desks; they're standing at packing stations, moving quickly, wearing gloves, and handling physical products. Our design adapts to them, not the other way around. Most importantly, we proved that real-time error detection isn't about replacing workers with automation, it's about empowering them with intelligent assistance that makes quality assurance continuous instead of reactive.

What we learned

We learned that domain expertise is everything. Understanding the Pick & Pack process; its physical constraints, timing pressures, and multi-airline complexity was critical to designing an effective solution. We learned that AI's role in operations should be assistive, not substitutive: operators remain in control, but now they have instant feedback and visual confirmation. We learned that data architecture defines scalability: good database modeling today enables intelligent systems tomorrow. The Product_AirlineAvailability table, for example, elegantly solves the multi-airline challenge without creating inventory chaos. We learned that user experience under pressure requires ruthless simplicity: large touch targets, minimal cognitive load, instant feedback, and mobile-first design aren't nice-to-haves; they're requirements when users work in high-speed, high-stakes environments. We learned that modern computer vision has crossed the threshold from experimental to operational: using AI for visual validation isn't a tech demo anymore; it's a practical solution that provides explainable results. Finally, we learned that in real-world development, execution matters more than ideas: when everyone has the same concept, the team that ships wins.

What's next for Gate Keeper

Gate Keeper's next evolution involves three critical paths. First, full backend integration: connecting the React frontend to a FastAPI service that handles Gemini AI vision API calls for real-time drawer validation, implements weight variance algorithms with configurable tolerance thresholds, and manages Snowflake database operations for all flight, spec, and validation data. Second, the supervisor dashboard: a live monitoring interface that shows all active packing stations in real-time, displays alert feeds by severity, visualizes operator performance metrics (time per drawer, error rates, rework frequency), and provides inventory visibility for Pick & Pack stock levels and replenishment needs. Third, operational intelligence: implementing analytics that identify systematic error patterns (which products, drawers, or shifts generate the most deviations), building predictive models for stock replenishment based on flight schedules, and creating compliance audit trails that link every drawer to its operator, timestamp, validation result, and any deviations. Beyond gategroup, the core concept, real-time validation at the point of action, is universally applicable: hospital pharmacies validating medication trays, warehouse fulfillment checking orders before shipping, food packaging ensuring compliance in regulated environments. Gate Keeper isn't just a hackathon project; it's a blueprint for digital transformation in any manual-intensive industry where human precision meets operational complexity and zero-defect tolerance.

Built With

• convexdb
• fastapi
• geminiapi
• nextjs
• python
• react
• tailwindcss
• typescript