EcoPrompt 🌱

An invisible layer that makes every AI conversation more efficient—automatically.

The Problem

Users write inefficient prompts on ChatGPT, Claude, and other AI platforms daily. Verbose, vague, or poorly-structured prompts waste computational resources, increase response times, and contribute to unnecessary carbon emissions at massive scale.

Example: "Hey, I was wondering if you could maybe help me understand some stuff about climate change and like what causes it and things like that?"

Our Solution

EcoPrompt is a Chrome extension that intercepts prompts before they reach the LLM, optimizes them using a lightweight model, and sends the efficient version—all invisibly in the background.

Users get the same answers, faster, while consuming significantly less energy.

How It Works

1. User types prompt into ChatGPT/Claude/etc.
2. User clicks submit
3. EcoPrompt intercepts the request
4. Lightweight optimizer (BERT-based or small SLM) rewrites the prompt to be concise and efficient
5. Optimized prompt is sent to the actual LLM
6. User receives response normally—faster and with less compute

All of this happens invisibly in milliseconds.

Why This Matters

Individual Impact

• 30-50% reduction in prompt tokens
• Faster responses (less text to process)
• Better results (clearer, more specific prompts)

Collective Impact

If 1 million users adopt EcoPrompt:

• Billions of tokens saved annually
• Tons of CO2 prevented from data center operations
• Meaningful reduction in AI's environmental footprint

Example Optimization

Before (67 tokens):

"Hey, I was wondering if you could maybe help me understand some stuff about climate change and like what causes it and the main things that contribute to it? I'm trying to learn more about this topic."

After (12 tokens):

"What are the top 5 causes of climate change?"

Savings: 82% token reduction, same information requested.

Current Implementation

For this hackathon proof-of-concept, we demonstrate the system using NVIDIA Nemotron Nano 9B V2—their smallest, most efficient model with reasoning budget controls.

Our vision: Replace this with a custom fine-tuned DistilBERT model (<50MB) for true local, instant optimization with zero API calls.

Supported Platforms

• ✅ ChatGPT (OpenAI)
• ✅ Claude (Anthropic)
• 🔄 Gemini, Perplexity (Coming soon)

Built With

NVIDIA Nemotron Nano 9B V2 • Chrome Extension APIs • Vercel

Built at NVIDIA Hackathon SF 2024

---

EcoPrompt: Invisible efficiency, monumental impact. Every prompt optimized, automatically.