Philadelphia has thousands of vacant, blighted properties and entire neighborhoods without reliable internet access. Both crises are invisible to the people who could fix them, because the data exists, but no one's made it actionable. Holmes changes that. It's a real-time civic intelligence platform with two lenses: a housing map that scores every vacant property for blight risk and tells you exactly where to intervene, and a connectivity map that surfaces the neighborhoods where broadband gaps are leaving residents behind. Point it at any neighborhood, and Holmes, backed by live city data and AI, tells you what's broken, why, and what to do about it. Built for Philadelphia. Built for the people trying to fix it.

Philadelphia has thousands of vacant, blighted properties and entire neighborhoods without reliable internet access. Both crises are well-documented and the data exists, but it's scattered across city databases and impossible to act on without the right tools to make sense of it. Holmes is a real-time civic intelligence platform built to close that gap. It ingests live data from OpenDataPhilly and Philadelphia's ArcGIS APIs, stores it in a Neon serverless Postgres database, and surfaces it through an interactive map and AI-driven analysis layer. The Housing Intelligence layer pulls property assessments, L&I code violations, and vacancy indicators and scores every blighted property on a 0-100 risk scale using weighted multi-factor logic. The interactive Leaflet map lets you drill into any property or neighborhood and trigger an AI brief that explains what's driving the risk and which city programs apply. Those briefs are generated by Cloudflare Workers AI running a Llama 3.1 model at the edge, with responses grounded in real data through a RAG pipeline backed by Pinecone. When a query comes in, Holmes embeds it, retrieves the most semantically relevant civic context from the vector store, and injects it into the prompt so every answer is anchored to actual Philadelphia policy, programs, and property records rather than hallucinated. The Dead Zone Detective layer maps broadband access gaps across Philadelphia's census tracts using household internet adoption rates, device availability counts, and income pressure indicators sourced from ArcGIS. Each tract gets a connectivity risk score. The same AI engine that handles housing analysis can break down any tract: who is underserved, what is driving it, and where a city agency or internet provider should focus first. An equity analysis layer then runs a bounding-box spatial join between connectivity tracts and nearby vacant properties and code violations to surface double-burden areas where housing stress and digital exclusion overlap. The Glass Box is an audit transparency dashboard. The data it displays is simulated, but the idea it demonstrates is real: every AI response Holmes generates could be checked by an ethical middleware layer before it reaches the user, and the Glass Box shows what that oversight looks like in practice. Any reviewer can inspect the full event log, what was triggered, what rule fired, what action was taken, and what the AI said before and after the intervention. The goal is that when Holmes eventually runs in production at scale, errors like a failed data ingestion, a hallucinated property detail, or a confidence threshold breach would surface here in plain language so that a city employee or community organizer with no technical background could catch them, flag them, and understand what happened. It turns AI accountability from an engineering concern into something anyone can read.

Next.js 16, React 19, Tailwind CSS v4, Framer Motion, Recharts, Leaflet, Cloudflare Workers AI, Neon Serverless Postgres, Pinecone, OpenNext, Pinecone RAG, Inhibitor API, OpenDataPhilly, ArcGIS, GitHub.

https://github.com/Maxlovinger/CodeFest

https://holmes.maxlovinger88.workers.dev