Clarity

An AI operations manager for small businesses that predicts rush hours, auto-schedules staff, and reorders inventory before any shortages happen.

Comment

Seals 🦭

An AI operations manager for small businesses that predicts rush hours, auto-schedules staff, and reorders inventory before shortages happen.

💡 Inspiration

Small business owners operate on razor-thin margins, especially in their first few years. One unexpectedly busy rush can overwhelm unprepared staff and hurt customer satisfaction, while over-ordering wastes money and spoils inventory.

We built Seals to keep business owners ahead of demand. By forecasting rush hours, optimizing staffing requirements, and automating supply orders, Seals reduces wasted labor costs, prevents stockouts, and simplifies day-to-day operations.

🎯 What it does

Rush Forecasting

Seals estimates how busy your store will be by combining:

  • Weather signals and historical foot traffic patterns across different weather conditions
  • Holiday data and previous rush patterns during those times
  • Public event data (concerts, gatherings, local surges)
  • Congestion and traffic patterns

Output: A busyness score (1-5) for today, tomorrow, and the week ahead.

Staffing Optimization + Scheduling

Seals tracks:

  • Observed customer flow
  • Expected demand from AI forecasts
  • Current employee schedules
  • Customer satisfaction during different periods

Using these signals, the system identifies:

  • Coverage gaps (rush hours with higher wait times and lower satisfaction)
  • Understaffed and overstaffed periods to optimize break times
  • Correlations between staffing decisions and customer satisfaction

The AI then:

  • Generates optimized schedules
  • Maintains necessary staffing for customer satisfaction without cutting into profit margins
  • Adapts to individual employee availability and preferred work times

Inventory Monitoring

Seals tracks inventory and adapts ordering behavior:

  • Adjusts reorder thresholds based on the busyness forecast
  • Considers menu/stock mix and upcoming demand signals
  • Enables one-click automatic order placement

🔧 How we built it

Architecture

Frontend: TailwindCSS + Cursor + OpenAI for UI development

Key Technical Decisions

Multi-Agent System: We split responsibilities across specialized agents (forecasting, staffing, inventory) to keep behavior streamlined and debuggable.

Semantic Retrieval: Implemented a vector database with semantic RAG using cosine similarity over 1,500+ nodes for pattern-based decision making instead of pure generative outputs.

Busyness Scoring: Created a simple 1-5 scale that instantly communicates predicted customer volume, eliminating guesswork for each shift.

Action Protocols: Built MCPs (Model Context Protocols) enabling the AI to trigger real-world actions:

  • Fetch real-time data from APIs
  • Make schedule changes
  • Place inventory orders

Data Sources

Map APIs & Event Data: Integrated public data for events, concerts, and gatherings, using historical traffic patterns to estimate customer influx.

Weather Tracking API: Monitors weather conditions and correlates them with historical foot traffic.

Customer Flow Tracking: Monitors actual vs. expected store traffic, employee count, and customer satisfaction to optimize staffing levels.

🚧 Challenges we ran into

Frontend/Backend Integration: Getting the UI to accurately reflect outputs from forecasting and staffing logic required significant iteration.

Design Constraints: Creating a polished, usable interface with limited time and resources.

API Integration: Coordinating multiple map APIs to work together seamlessly taught us the importance of planning integration from the start.

Prompt Engineering: We learned the hard way that AI agents need hyper-specific constraints. Vague prompts led to broken frontends and 30-minute debugging sessions.

UI Refinement: Small styling changes (like adjusting a single circle's color) consumed disproportionate amounts of time.

🏆 Accomplishments that we're proud of

It works! Real-time data + AI agents = functional management optimization.

Semantic RAG Implementation: Built a vector database using cosine similarity over 1,500+ nodes to predict foot traffic metrics.

Model Context Protocols: Created MCPs enabling the system to:

  • Place supply orders
  • Schedule employees
  • Send notifications

Polished UI: Delivered a surprisingly functional and attractive web interface given our time constraints.

Multi-Agent Orchestration: Successfully coordinated multiple LLMs across various APIs to work cohesively.

📚 What we learned

Multi-Agent Orchestration: How to coordinate different AI agents across multiple APIs while maintaining useful, coherent outputs.

RAG Design: Designed and implemented our first semantic retrieval system from scratch.

Working Under Constraints: Learned to stay resourceful and make smart tradeoffs when time and resources are limited.

Data Integration: Converted disparate public data sources into immediately actionable insights for business owners.

Frontend/Backend Alignment: Improved our ability to create seamless integrations between UI and backend logic.

API Mastery: Gained hands-on experience with TomTom API and Open-Meteo.

🚀 What's next for Seals

Expand Business Models: Adapt the platform for bodegas, grocery stores, and other brick-and-mortar retail environments beyond restaurants.

Enhanced Forecasting: Integrate deeper demand signals like POS (point-of-sale) data to improve prediction accuracy.

Real-World Validation: Partner with small businesses to test and refine the system in production environments.

Mobile App: Develop a mobile-first experience for on-the-go business owners.

Built with ❤️ by the Seals team

Built With

Share this project:

Updates