Shift: personalized sustainability with radical transparency about AI's energy cost

By Suryansh Sijwali · · 2 min read · Engineering, Hackathon, Sustainability

The premise

65% of people want to live more sustainably. Only 26% follow through. The gap is not motivation, it is tools. Existing apps show guilt dashboards and generic tip lists. They don’t tell you what to do today, in your city, with your diet, on your commute.

And every AI product quietly ignores its own environmental footprint, which makes the irony complete: using AI to save the planet while burning energy to do it.

Shift addresses both halves.

The personalization half

Onboards in 90 seconds. Then delivers one AI-personalized sustainability micro-action per day, tailored to:

  • Your commute distance
  • Your diet pattern
  • Your city’s live grid carbon intensity
  • Current weather

Actions are grounded in EPA and DEFRA emissions data, structured using behavioral science frameworks (Fogg’s B=MAP, Tiny Habits), and scored against a curated knowledge base of 190 actions. Users earn points, build streaks, advance through five levels, and track contributions to UN SDGs.

The transparency half

Unlike every other AI product, Shift shows you what the AI costs:

  • Every action card displays the inference carbon cost alongside the savings the action enables.
  • A Chrome extension monitors environmental impact of every Gemini prompt in real time.
  • A dedicated Eco-LLM dashboard tracks energy (Wh), carbon (gCO2), and water (mL) per query.
  • Semantic caching serves similar queries without extra inference, so repeated questions cost zero.

Typical carbon ROI: 10,000 to 1 or higher (the action’s impact dwarfs the inference cost). The point isn’t that AI is free; it’s that the math should be visible.

Stack

Next.js 14 (App Router, PWA), TypeScript, Tailwind, shadcn/ui, Framer Motion, Tremor. Groq (Llama 3.3-70B) with Gemini fallback via Vercel AI SDK. Supabase (Postgres + pgvector). Upstash Redis (TTL cache) + Upstash Vector (semantic dedup). Climatiq, Electricity Maps, Google Maps Distance Matrix, Open-Meteo. EcoLogits for carbon estimation with a Groq LPU efficiency multiplier. PostHog + Sentry. Resend for email.

What this could be

At 100,000 daily users completing one action each, the rough math is 12,000 tonnes of CO2 per year removed (about 2,600 cars). The Eco-LLM transparency layer is also a standalone product for enterprise teams navigating AI carbon disclosure under EU AI Act regulations.

Built for

GDG @ Penn State Solution Challenge, 12 hours, three people: Suryansh Sijwali, Nabeel Ahmed, Neil Barbara. Climate Change Track winner.

Receives occasional maintenance from the team.