CoffeeScout

Turn real-world mobility data into smarter expansion decisions. CoffeeScout helps businesses discover promising locations for their next coffee shop.

Comment

Inspiration

Choosing the right location is one of the most important decisions when expanding a physical business. A great product can still struggle if the location does not have enough activity or the right movement patterns.

We wanted to build something that makes location analysis easier for a business owner without requiring them to understand SQL, data analytics, or complex datasets.

That idea led to CoffeeScout — a location intelligence platform that helps businesses discover promising locations for their next coffee shop using real-world mobility data and AI.

What CoffeeScout Does

Instead of simply showing where existing coffee shops are, CoffeeScout focuses on a different question:

"Where could my next coffee shop be worth considering?"

A user can enter a natural-language request such as:

"Find the best coffee shop locations in Brooklyn based on Citi Bike activity."

CoffeeScout then analyzes the request, retrieves relevant real-world data, identifies promising locations, and presents the results visually on an interactive map.

Users can then:

  • Explore recommended locations on a map
  • Understand why a location was identified
  • View the underlying data signals
  • Compare candidate locations
  • Save promising locations
  • Share a location recommendation
  • Open the location directly in Maps

How We Built It

CoffeeScout is built as a React + FastAPI application.

The core architecture is:

React → FastAPI → Google ADK → Gemma 4 → BigQuery MCP Server → BigQuery

The user interacts with the React frontend through a natural-language search.

Google ADK orchestrates the agent workflow, while Gemma 4 helps understand the user's intent and interpret the retrieved information.

Through the BigQuery MCP Server, the agent can interact with BigQuery and retrieve data from the public NYC Citi Bike dataset.

The returned station activity and location data are then used to identify candidate locations and visualize them using real coordinates on the map.

The important part is that CoffeeScout does not simply generate a location from the model's memory. The recommendation is grounded in data retrieved from BigQuery.

Why This Is Different

Traditional map applications help users find places that already exist.

CoffeeScout focuses on a different problem:

Helping a business think about where its next location could be.

The combination of natural-language interaction, an AI agent, MCP-based data access, real-world mobility data, and explainable location insights turns a complex data-analysis workflow into a simple product experience.

Although our current demonstration focuses on coffee-shop expansion, the underlying location-intelligence approach could eventually be adapted to restaurants, retail stores, gyms, EV charging stations, and other location-based businesses.

Challenges We Faced

One of the biggest challenges was making sure the AI-generated recommendations were actually connected to real data rather than static or fabricated results.

We also had to solve query-specific location analysis. For example, a request for Manhattan should not simply return the same results as a request for Brooklyn or Queens.

Another challenge was connecting the AI agent, BigQuery MCP Server, backend APIs, and interactive React map into a single smooth workflow.

We also focused on making the final experience feel like a real SaaS product rather than a simple AI demo.

What We Learned

Building CoffeeScout taught us how to combine several technologies into an agentic data workflow.

We learned how to:

  • Build an AI agent using Google ADK
  • Use Gemma 4 for natural-language reasoning
  • Connect an AI agent to external data through MCP
  • Query and work with BigQuery public datasets
  • Build a React + FastAPI architecture
  • Transform data into interactive map-based insights
  • Design AI recommendations around real data signals
  • Handle the difference between AI interpretation and factual data

Most importantly, we learned that an AI application becomes much more useful when the model can interact with real-world data instead of operating only as a conversational interface.

Impact

CoffeeScout aims to make location analysis more accessible to small businesses and entrepreneurs who may not have dedicated data teams.

Instead of spending hours manually exploring datasets, a business owner can start with a simple question and receive a data-backed, explainable view of potential opportunities.

CoffeeScout turns a complex location-analysis problem into a simple search, analysis, and discovery experience.

Built With

Share this project:

Updates