Eco-Health Schedule Agent - Willow

Willow personalizes and optimizes daily routines, outdoor activities, and home ventilation around real-time atmospheric metrics—such as Air Quality (AQI), UV index, and pollen levels.

Comment

Inspiration

With recent wildfires causing frequent air quality spikes and a growing number of people managing respiratory sensitivities like asthma and pollen allergies, staying safe outdoors has become increasingly tricky. I wanted to build an intuitive, conversational assistant that bridges the gap between complex meteorological data and everyday life—making it effortless for anyone to ask quick questions and protect their health without having to decipher raw data charts.

What It Does

Willow is an AI scheduling assistant built to optimize daily routines, outdoor activities, and home ventilation around real-time atmospheric metrics—such as Air Quality (AQI), UV index, and specific pollen levels—to protect individual health during optimal environmental windows.

Willow learns your personal profile (such as age, location, and specific sensitivities like ragweed or asthma) and stores it across sessions. The agent proactively evaluates hourly environmental forecasts to reschedule risky outdoor workouts, suggest ideal times to open windows for ventilation, and tailor safety warnings directly to your personal health needs.

How I Built It

  • Google Agent Development Kit (ADK): Orchestrated multi-turn conversational flows, system instructions, and tool execution.
  • Gemini (3.5 Flash/3.1 Flash-Lite): Powered the core reasoning engine for schedule optimization, context handling, and natural language understanding.
  • Custom Tool Integrations & APIs: Integrated real-time atmospheric endpoints (covering hourly AQI, PM2.5, UV index, and pollen counts) and built state persistence mechanisms using ADK session state management.
  • Python: Handled backend logic, API request pipelines, and custom agent tool interfaces.

Challenges I Ran Into

  • Data Integration & API Latency: Aggregating multiple environmental metrics (air pollutants, pollen types, and UV ratings) into a single, cohesive schema that Gemini could parse and evaluate in real time without lagging.
  • Session State Management: Early on, the agent struggled to persist user health profiles and would freeze or fail to respond after executing state-saving tools. Debugging ADK context injection, tool argument handling, and session state lifecycles was a major hurdle that required refining how the model triggered functions and parsed persistent variables.

Accomplishments That We're Proud Of

  • Seamless Personalization: Building a reliable, natural onboarding experience where the agent remembers personal allergies and health contexts across interactions without feeling intrusive.
  • Smart Decision-Making: Seeing the agent accurately evaluate conflicting variables (e.g., high heat vs. high pollen vs. clean AQI) and intelligently propose optimized schedule swaps that genuinely help users protect their health.
  • End-to-End Execution: Delivering a fully functional, reliable assistant from idea to working prototype in just a few days.

What We Learned

  • How to design, build, and debug agents using the Google Agent Development Kit (ADK) and Gemini's tool-calling ecosystem.
  • How to structure user-scoped memory and state management in conversational AI workflows.
  • Best practices for managing an independent end-to-end repository, from virtual environments to clean modular code architecture.

What's Next for Willow

  • Sustainability & Eco-Action Module: Expanding Willow with a waste-reduction assistant that gives instant guidance on responsible recycling, composting, e-waste drop-offs, and upcycling.
  • Personal Carbon Footprint Tracker: Adding tools to estimate daily transit emissions and suggest lower-impact alternatives for regular routines.

Built With

Share this project:

Updates