Stravify - Ask your running activities

Stravify is an AI-powered analytics dashboard that allows runners to chat with their Strava data using natural language. Built with a modern React frontend and a robust FastAPI backend, it leverages Google's Gemini models via LangChain to analyze your running history dynamically.

🚀 Features

• Natural Language Querying: Ask questions like "How many miles did I run last week?" or "What was my average pace in December?".
• Real-time Analysis: Fetches your latest Strava activities on the fly.
• Secure Authentication: Uses Strava OAuth2 for secure user login.
• Modern UI: A dark-themed, responsive interface built with Tailwind CSS.

🛠️ Technical Stack

Frontend

• React + Vite: Fast, component-based UI development.
• Tailwind CSS: Utility-first styling for a sleek, dark-mode design.
• Lucide React: Beautiful, consistent iconography.
• Axios: For handling API requests to the backend.

Backend

• FastAPI: High-performance Python web framework for handling API requests and async operations.
• LangChain: Orchestrates the AI interactions and tool usage.
• Google Gemini (gemini-2.5-flash): The underlying LLM that powers the analysis and code generation.
• Pandas: Used for efficient data manipulation and analysis of running activities.
• Supabase: (Optional) For persistent storage of chat logs and user profiles.

🤖 How AI is Used

Stravify doesn't just "guess" answers. It uses a ReAct (Reasoning + Acting) agent workflow:

1. Data Loading: When you log in, the backend fetches your last 200 activities from Strava.
2. DataFrame Creation: These activities are converted into a structured Pandas DataFrame containing metrics like distance, moving time, elevation, and dates.
3. Agent Execution:
   • When you ask a question, the LangChain Pandas DataFrame Agent receives the query.
   • The agent uses Google Gemini 2.5 Flash to reason about the data.
   • It generates Python code (specifically Pandas operations) to calculate the exact answer.
   • It executes this code in a secure sandboxed environment.
   • Finally, it interprets the result and returns a friendly, human-readable response.

This approach ensures 100% mathematical accuracy for queries like sums and averages, as the AI isn't doing mental math but rather writing code to solve the problem.

📦 Setup & Installation

Prerequisites

• Node.js & npm
• Python 3.10+
• Strava API Credentials
• Google AI Studio API Key

Backend Setup

1. Navigate to backend/:

   cd backend

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Configure .env:

   STRAVA_CLIENT_ID=your_id
   STRAVA_CLIENT_SECRET=your_secret
   GOOGLE_API_KEY=your_gemini_key

5. Run the server:

   uvicorn main:app --reload

Frontend Setup

1. Navigate to frontend/:

   cd frontend

2. Install dependencies:

   npm install

3. Run the development server:

   npm run dev