SustainAI

Problem

Sustainability tools tend to fail in one of two directions. Either they report a global average that tells you nothing about where you live, or they present a confident local number that the underlying data cannot actually support — a neighbourhood air-quality reading derived from a 45km model grid, say. Both leave a user unable to judge how much to trust what they are seeing.

• Global figures are too coarse to act on
• Precise-looking local figures are often modelled, not measured
• Without knowing which is which, a user cannot calibrate their trust

Solution

A location-first workspace: pick a place, see what is genuinely known about it, and model changes against your own measured baseline rather than someone else's assumptions.

• Dashboard — temperature, air quality, a 7-day rainfall outlook and active goals for the selected place
• Interactive map — search a city, address or coordinates, or click anywhere to analyse that point
• Predictions — 7-day weather forecasts and regional air-quality forecasts
• What-If — enter monthly water, electricity and car use, then move reduction sliders to estimate savings, with optional carbon factors
• Goals — measurable targets with a category, unit and deadline, plus progress history
• Action plans — save suggested actions or scenarios and mark them complete
• Saved locations and accounts, so a signed-in user's places and goals stay private to them

Transparency as a feature

The Data Sources page is the part of this project worth reading first. It lists every provider with its coverage, resolution, data type and timestamp, and then states outright what the app cannot conclude: there is no worldwide neighbourhood-level feed for water consumption, electricity mix, traffic or biodiversity, so those metrics are simply absent rather than estimated. It notes that a regional forecast is never presented as a live local sensor, that the proxy score uses US AQI only, and that the cross-system connections shown elsewhere in the app are explanatory hypotheses rather than locally measured effects.

• Each source carries its coverage, resolution and retrieval time
• Provider timestamps distinguish modelled time from model issuance
• Scenario results are labelled as dependent on the user's own inputs
• ODbL and Open-Meteo attribution and terms are linked, not buried

Data pipeline

Six stages, shown in the app itself: search, geocode, fetch verified data, analyse, predict, recommend. Nothing is synthesised between those steps — where a provider has no data, the interface shows an empty state rather than a plausible number.

• Photon / OpenStreetMap for place and address search
• Open-Meteo for current conditions and 7-day forecasts, on ~1–25km model grids
• CAMS via Open-Meteo for air quality, on a global 0.4° (~45km) grid
• The user's own measured baseline for anything about their household

Status

Live and publicly usable at eco.srbros.in, hosted on Cloudflare — no sign-in needed to explore a location. Username-and-password accounts keep saved locations, goals and plans private; Google sign-in is being prepared. The AI assistant requires a working AI service key to activate. Development is ongoing.