FacEmotion

problem

Detect facial emotions in real time and recommend personalized actions to improve user mood.

approach

CNN optimized on a 35K+ image dataset with robust preprocessing and augmentation; Gemini 1.5 Pro integration for context-aware recommendations; Streamlit + OpenCV for low-latency delivery.

implementation notes

• Built a facial-emotion analysis system that secured first place among 53 teams in the UCSB Data Science Competition.
• Designed and trained a CNN architecture on 35,000+ labeled facial images, achieving 87% accuracy in multi-class emotion detection.
• Integrated Gemini 1.5 Pro to provide personalized, real-time mood enhancement recommendations based on live frame analysis.
• Built an end-to-end prototype with OpenCV preprocessing and a Streamlit UI for responsive inference and user feedback.
• Outcomes: 1st Place Winner — UCSB Data Science Competition; 87% emotion detection accuracy; 35,000+ training images.

impact

recognition

• 1st place among 53 teams — UCSB Data Science Competition

links

• github.com