optimizeAI

cuhackit25 tower janitors • Reduced LLM compute costs by up to 30% through a multi-stage query routing system that combined semantic similarity, web search, and cached responses to divert non-complex requests away from expensive models • Built a Next.js analytics dashboard that ingested model usage logs and energy metrics, computing real-time CO2 savings trends through interactive visualizations to guide optimization of query routing and model selection • Developed a gamified rewards engine using token-based incentives and progress tracking to encourage sustainable behavior, boosting long-term user retention and engagement across repeated platform session