EcoPrompt

Engineer. USC AI4GOOD Hackathon, April 2026.

Next.js, TypeScript, AWS Bedrock, Titan embeddings, DynamoDB, Tailwind CSS

Code on GitHub

I built a chat app that reuses answers to similar questions, sends simpler prompts to a smaller model, and shows the modeled energy saved.

Less modeled energy per query on a cache hit, not metered: 98.6%
Less modeled energy per query on a right-sized model, not metered: 90%

Before making another model call

Titan embeddings let the app look for a similar question in DynamoDB and reuse its answer. For a new question, prompt complexity picks between Claude Haiku and Sonnet.

A model, not a meter

I used an energy model because of the hackathon clock. The 98.6% cache-hit saving comes from assumed costs of 0.0001 against 0.007 kWh per query, and the 90% right-sizing saving uses 0.0007 against 0.007. EPA eGRID factors turn that energy into carbon. None of it is a hardware measurement.

Who built it

I did the engineering and worked with one designer: the chat interface, semantic caching, model selection, and the energy dashboard.

Next: Converse