Sustain-A-Prompt

HomePage

"""## Inspiration

Generative AI is powerful but energy-intensive. We wanted a way to make “green by default” prompting tangible: show real energy, CO₂, and water impacts for everyday prompts and help people reduce them without sacrificing quality.
Teams asked for something practical they can adopt today—fast feedback, model-aware guidance, and a shared language (clear equivalents like homes powered, km driven, and daily drinking water).

What it does

Analyzes any prompt and estimates its environmental footprint across energy (kWh), CO₂ (g), and water (L), scaled to a 10,000,000-token scenario for easy sense-making.
Optimizes your prompt and shows token savings with a side-by-side compare.
Model-aware: pick a base model to see footprint and cost deltas that reflect that model’s characteristics.
Adds context with relatable legends: homes powered for a day (energy), people’s drinking water for a day (water), and km driven in a typical car (CO₂).
Gamification & education: earn badges, view a real-time globe of activity, and learn how HPC training vs. inference impacts sustainability.

How we built it

Frontend: Next.js 15 (App Router) + TypeScript with shadcn/ui and Tailwind for a dark, neon, glassmorphism aesthetic; framer-motion for smooth pipeline and UI animations; lucide-react for icons.
Components:

Prompt console with a pipeline animation (tokenizing → processing → generating → complete).
Impact cards that show energy, CO₂, and water with model-aware scaling to 10M tokens and clear legends.
Optimized Prompt compare panel with savings.
Model Picker with live deltas based on per-token metrics and estimated costs.
3D Globe for global activity and an HPC education card for context.

Backend:

Serverless API routes that call external LLMs (Gemini optimize and Cerebras chat).
Robust fallback estimation when APIs are unavailable: local per-token model data → consistent scaling → user-friendly legends.

UX:

Tron-like neon theme, glass panels, and subtle motion to keep it inspiring yet focused.
Accessibility considerations and responsive layout.

Challenges we ran into

Making sustainability metrics understandable without oversimplifying: picking consistent baselines (homes/day kWh, drinking water/day, g CO₂/km) and scaling to 10M tokens.
Ensuring model-aware accuracy while providing graceful fallbacks when token usage isn’t returned by providers.
Balancing performance (3D globe, animations) with clarity on low-powered devices.
Designing a clean flow that fits both newcomers and advanced users without overwhelming them.

Accomplishments that we're proud of

A cohesive, production-ready experience that turns abstract sustainability into clear, relatable insights.
Model-aware impact computation with consistent 10M-token scaling and human-friendly legends.
A delightful UI—glassmorphism, neon accents, subtle motion—that still feels fast and accessible.
Gamification that nudges better behavior (badges, progress) without being intrusive.
An education layer (HPC card) that empowers users to make informed choices beyond just “use fewer tokens.”

What we learned

“Green prompting” resonates when you connect it to everyday equivalents and give actionable next steps.
Small prompt improvements can compound massively at scale; the 10M-token lens makes this obvious.
Reliability matters: people trust tools that handle API hiccups gracefully and still provide meaningful estimates.
Visual feedback (pipeline, globe) increases engagement and understanding without heavy text.

What's next for Sustain-A-Prompt

Per-metric savings in the Optimized Prompt compare (energy/CO₂/water bars with their own percentages).
Team features: shared workspaces, leaderboards, and policy presets (e.g., “always prefer efficient models”).
Reports and exports (CSV/PDF) for sustainability teams and procurement.
Browser extension and IDE plugins to optimize prompts in-context.
More providers/models and periodic calibration against independent benchmarks.
Deeper education modules (e.g., region-specific electricity mixes, renewable offsets) and richer badges. """