# UrbanVitals | Sherwin Vishesh Jathanna > ๐ŸŽ–๏ธ Honorable Mentions at HackMIT 2025 (Tandemn: Best Use of AI) & ๐Ÿ† Best Student Hack at StarHacks 2025. UrbanVitals is an AI-powered sustainability platformโ€ฆ Source: https://www.sherwinvishesh.com/project/urban-vitals Copyright 2026 Sherwin Vishesh Jathanna. Text may be quoted with attribution. The design and source code are not licensed for reuse. ๐ŸŽ–๏ธ Honorable Mentions at HackMIT 2025 (Tandemn: Best Use of AI) & ๐Ÿ† Best Student Hack at StarHacks 2025. UrbanVitals is an AI-powered sustainability platform that evaluates neighborhoods using comprehensive environmental, infrastructure, and livability metrics. Everyone talks about building greener cities, but if you actually ask *how sustainable is my neighborhood*, there is rarely a straight answer. The data exists, scattered across air-quality feeds, energy grids, transit maps, and government reports, yet it is far too fragmented for a regular person, or even a city planner, to make sense of. **UrbanVitals** pulls all of that together into a single, honest number. It is an *AI-powered sustainability platform* that scores the livability and environmental health of urban neighborhoods, and it picked up two awards along the way: an **Honorable Mention for Best Use of AI (Tandemn) at HackMIT 2025** and **Best Student Hack at StarHacks 2025**. ### One number for a neighborhood's health At the center of UrbanVitals is the **Green Score**, a rating from *1 to 10* built from more than *13 indicators* spanning the environmental, infrastructural, and social sides of a place: air and water quality, renewable energy use, grid reliability, transit access, walkability, and more. Under the hood, a **weighted model** groups those signals into categories like *Environmental Quality*, *Infrastructure*, *Livability*, *Sustainability*, and *Disaster Risk*, then blends them into one figure you can compare across neighborhoods at a glance. ### A 3D map you can actually read The frontend, built with **React** and **Mapbox GL**, turns all of that into an interactive *3D map*. Neighborhoods are color-coded by their Green Score, and you can toggle layers to surface specific concerns like greenery density, pollution hotspots, or disaster vulnerability. Smooth transitions and hover details make the whole thing feel exploratory rather than clinical, so someone with no technical background can dig into their own street as easily as an urban planner can. ### Grounded in real data, explained in plain language A score is only as trustworthy as what feeds it. On the backend, a **FastAPI and Python** service, accelerated by the **Cerebras SDK** for the heavy AI work, runs a pipeline that consolidates open data from **OpenStreetMap**, **Open-Meteo**, and government environmental APIs, so every Green Score is grounded in *verifiable, real-world data* rather than guesswork. To explain the results, the AI uses **Retrieval-Augmented Generation** with **LangChain**, pulling in relevant sustainability context before it answers, which keeps its explanations accurate and *traceable*. A built-in conversational assistant lets you ask why a neighborhood scored the way it did, compare places, and get concrete improvement ideas, and every score stays explainable: a low mark might trace straight back to poor recycling rates or an unreliable power grid, shown right there in the dashboard. ### AI that practices what it preaches The part I am proudest of is that UrbanVitals does not just *measure* sustainability, it *lives* it. The AI runs on **Tandemn**, a sustainable computing platform that uses refurbished hardware powered by renewable energy, which saves roughly **1.7 mg of CO2 per token** the model generates. Every query in the app tracks and displays its own *real-time carbon savings*, so the tool is honest about its own footprint even while it analyzes everyone else's. ### Built for a city, designed to scale For the hackathon, the team, **Divyam Kataria**, **Vivien Lim**, and I, built the MVP around **Tempe, Arizona**, aggregating live environmental and infrastructure data and scoring it in real time. The architecture is deliberately modular: it supports **Modal cloud** deployment so large citywide analyses can run in parallel, with **60fps Mapbox rendering** and **Redis caching** keeping it fast even across thousands of data points. From here, the vision reaches toward multi-city and international coverage, predictive climate modeling, urban-planning simulations, and policy recommendation tools, turning UrbanVitals from a scorecard into a genuine decision-making aid for building better cities. ## At a glance - **Duration:** 48 hours - **Type:** Hackathon Project - **Event:** StarHacks 2025 & HackMIT 2025 - **Award:** Best Student Hack & Honorable Mentions - Tandemn: Best Use of AI - **Other Contributors:** Divyam Kataria, Vivien Lim ## Technologies - React.js - Vite - FastAPI - Python - Mapbox GL JS - Cerebras SDK - Tandemn AI - LangChain - Retrieval-Augmented Generation (RAG) - OpenStreetMap & Open-Meteo APIs - Modal Cloud - Redis (Caching) - Data Visualization (3D Mapping) ## Categories - Web App - AI/ML - Hackathon Submission - Sustainability ## Photos - Demoing UrbanVitals with the team, the neighborhood map live on the laptops - With the team and Ali Dasdan, CTO of Dropbox, at HackMIT 2025 - In front of MIT's Great Dome at HackMIT ## Links - [GitHub](https://github.com/sherwinvishesh/Urban-Vitals) - [Devpost](https://devpost.com/software/urban-vitals) - [Plume(HackMIT)](https://plume.hackmit.org/project/zmzpo-qcxdh-scgfu-uwpnh) - [Demo video](https://www.youtube.com/watch?v=GOx4Svgmur8)