Multimodal Agentic AI · May 2025 - Jun 2025
EcoMate-AI
A multimodal agentic platform that turns receipts, images, and everyday activity into personalized sustainability actions - designed, built, and demoed in 24 hours.
1st place
UCLA SAIRS Hackathon, judged by Microsoft, IBM, NVIDIA, Google, LinkedIn
24 hours
Concept to working demo
The problem
Most people have no idea what their daily choices cost the planet, and sustainability tools demand tedious manual entry. The challenge: make environmental impact effortless to understand from the messy data people already have - receipts, photos, and activity.
What I built
- A multimodal agentic platform on Gemini 2.5 Flash that ingests text, images, and structured data.
- Agent workflows that map inputs to impact and generate personalized sustainability recommendations.
- A live, demo-ready app shipped within the 24-hour hackathon window.
Outcome
- Won 1st place in the sustainability track against an industry judging panel.
- Validated a fast multimodal-to-recommendation pipeline that turns raw inputs into actions.
Tech
Gemini 2.5 FlashPythonStreamlitComputer VisionNLPRAG
Full project report
A detailed write-up for this project is coming soon.