ApexTracker AI 🏅

A Context-Aware Hybrid Training Agent built for the Gen AI Academy (APAC Edition).

ApexTracker AI is an elite, intelligent sports science coach designed for athletes tackling complex, multi-disciplinary events. By combining real-time environmental data with a vector-searchable sports science database, it dynamically adjusts daily training loads, recovery protocols, and pacing strategies.

---

🖥️ How to Use the Web App

Using the live agent is simple and requires just two inputs:

1. Open the Live Demo: Click the link above to access the Streamlit web interface.
2. Enter Your Location: In the "Training Location" box, type your current city (e.g., Bengaluru). The agent will use this to fetch live weather and humidity data via WeatherAPI.
3. Log Your Status/Query: In the text area, describe your planned workout, your specific race goals, or any fatigue/pain you are experiencing. (e.g., "I have a 10km run today, but my shins hurt.")
4. Consult the AI: Click the "Consult AI Coach" button.
5. Review Your Plan: The app will display the live data it gathered (Weather + Database protocols) and generate a custom, safe, and actionable daily training plan tailored to your exact situation.

---

🏆 Tracks

This project was built to address all three core tracks of the Gen AI Academy:

1. Track 1: Build and Deploy AI Agent: A fully autonomous conversational agent powered by Google Gemini 1.5 Flash, wrapped in a Streamlit frontend, and deployed via Docker to Google Cloud Run.
2. Track 2: Connect AI Agents to Real-World Data: Integrated a custom Model Context Protocol (MCP) tool utilizing WeatherAPI to fetch live, local environmental conditions to inform training safety.
3. Track 3: Build and Migrate faster with AI-Ready Databases: Utilized Google Cloud AlloyDB for PostgreSQL with the pgvector extension to store, embed, and retrieve highly specific sports science recovery and pacing protocols.

---

🎯 Example Prompts to Try

To see the full power of the agent, try copying and pasting these scenarios into the query box!

Example 1: Acute Recovery & Injury Management (Tests Track 2 & 3)

• Location: Bengaluru
• Your Query: "I'm experiencing heavy quad cramps and some shin pain after my last half marathon. I had a 10km run scheduled for today. What should I do?"
• What to Expect: The AI will instruct you to immediately reduce running volume by 50%, suggest substituting the run with low-impact swimming to maintain your aerobic base, and provide specific hydration protocols from the AlloyDB database.

Example 2: Event Pacing & Strategy (Tests Track 3)

• Location: London
• Your Query: "I'm training for an upcoming Hyrox match on April 11. My goal is a 1 hour and 30-minute finish time. How should I approach my simulated workout today?"
• What to Expect: The AI will pull the specific strategy from the vector database, advising you to maintain a strict Zone 3 heart rate on the 1km runs, bank your time on the sled push and row, and recover actively during the farmer's carry.

Example 3: Environmental Adaptation (Tests Track 2)

• Location: Dubai
• Your Query: "I'm feeling good and want to do an outdoor threshold run today."
• What to Expect: Because the MCP tool will detect extreme heat in Dubai, the AI will intercept the plan and dynamically suggest an indoor alternative, like a treadmill session or an indoor pool workout, to prevent heat exhaustion.

---

⚙️ Tech Stack

• Frontend: Streamlit (Python)
• AI Model: Google Gemini 2.5 Flash (google-generativeai)
• Database: Google Cloud AlloyDB (psycopg2-binary, pgvector)
• External APIs: WeatherAPI
• Deployment: Docker, Google Cloud Run

---

📂 Project Structure

apextracker/
│
├── app.py               # Core Streamlit application and Gemini agent logic
├── database_init.py     # Script to initialize AlloyDB and populate recovery protocols
├── weather_tool.py      # Tool to fetch real-time environmental data
├── requirements.txt     # Python dependencies
└── Dockerfile           # Containerization instructions for Cloud Run