This is a submission for the Built with Google Gemini: Writing Challenge
What I Built with Google Gemini
GreenAI-Agent is an intelligent assistant designed to help developers and organizations monitor and optimize the carbon footprint of their AI workloads. As AI models become more complex, their energy consumption grows; this project solves the "visibility gap" by providing real-time insights into the environmental impact of code execution.
Google Gemini served as the "brain" of the agent. I used it to:
- Analyze complex energy usage logs and translate raw data into actionable "green" recommendations.
- Power the natural language interface, allowing users to ask questions like "Which of my functions is consuming the most power?"
- Generate optimized code snippets that prioritize efficiency without sacrificing performance.
Demo
You can find the full source code and documentation here:
GitHub: https://github.com/git791/GreenAI-Agent
StreamLit: https://greenai-agent-pgttexvww7m2bpeschkc6n.streamlit.app/
What I Learned
Building this project taught me a lot about sustainable computing and the nuance of LLM token efficiency.
- Technical: I sharpened my skills in prompt engineering—specifically how to ground Gemini's responses in specific hardware telemetry data.
- Unexpected Lesson: I realized that even the "Green Agent" has a footprint! It led me to implement a "low-power" mode for the agent itself, where it uses more concise prompts to save tokens and energy.
Google Gemini Feedback
- The Good: The context window is a lifesaver. Being able to feed in large logs of execution data without the model "forgetting" the beginning of the run made the analysis incredibly accurate.
- The Friction: I ran into some challenges with rate limiting when trying to do high-frequency real-time monitoring. I had to implement a batching system to send data to Gemini every 30 seconds rather than instantly to stay within the free tier limits during development.
Top comments (0)