Hero Landing Page
Overview of the Mapping GUI
Overview of a Development POI
Indepth Development Info
Data for an Amenity POI
Custom Commands for interfacing

MapD – Urban Development Intelligence

Inspiration

We all see the signs: a new condo, a townhouse complex, a high-rise. Most people walk by with a lingering question: How will this actually impact our community during construction and long after? MapD turns that concern into clarity.

What it does

MapD is a geospatial storytelling platform that translates fragmented open data into a single, understandable view of an urban project. Drop a pin (or paste a permit address) and MapD:

Aggregates relevant public datasets (permits, zoning, transit, 311/service requests, and more).
Normalizes & links records to the same place and time windows.
Summarizes likely impacts (traffic, noise, accessibility, nearby amenities) into plain language.
Visualizes changes with a clean, minimal map.

How we built it

24-hour goals

Ship a deployed MVP
Prioritize unity & clarity over flash
Engage our community with approachable language

Frontend: Vite + React + Tailwind for a unified, minimalist UI/UX (Mapbox GL for interactive layers).
Backend: Python + FastAPI with MongoDB for schema-light ingestion, all containerized via Docker Compose.
Intelligence: OpenAI-assisted parsing to reconcile messy fields (addresses, parcel IDs, neighborhood names) and to generate concise, non-technical summaries.
Infra: Vercel (frontend) + DigitalOcean (API) + MongoDB, with Cloudflare Workers bridging edge routing and caching.

Challenges we ran into

No common standards across open datasets. Address formats, IDs, and time fields varied wildly. We built a reconciliation layer (geocoding + fuzzy matching) to align records.
Concurrent cleaning and embedding. We needed embeddings early for entity linking while cleansing was still in flight; we solved this with staged, idempotent jobs.
Token efficiency at scale. Summarizing many small records can be costly; we batched by spatial/temporal tiles and cached intermediate prompts.
Mapbox GL integration details. Style performance and hit-testing across overlapping layers required careful source/layer ordering and debounced queries.

Accomplishments that we’re proud of

Deployed MVP that lets anyone paste an address and see a unified, human-readable impact brief in seconds.
One-click summaries: auto-generated “What this means for your block” blurbs with cited data rows.
Deterministic ETL → LLM loop: prompts are fed only clean, structured fields; results are cached with traceable inputs.
Graceful UX under messy data: if a dataset is missing, the UI explains what’s absent and why—no silent failures.

What we learned

Clarity beats complexity. People want “what, where, when, why it matters” in <200 words, backed by a map.
Spatial joins need human language. A parcel ID is useless to a resident unless translated into landmarks and street names.
LLMs excel as glue, after rigor. With consistent schemas, small prompts do more, cheaper.
Performance is UX. Caching and debouncing map events mattered as much as typography.

What’s next for MapD

More data sources: development phases, construction schedules, bike/ped safety layers, air/noise sensors.
Community notes: let residents pin local knowledge (detours, school pick-up congestion) with light moderation.
Impact scenarios: compare “as-is” vs. “post-build” (parking load, stop frequency, service requests).
Accessibility & language support: alt-text for all map layers and summaries in multiple languages.
Civic partner pilot: work with a municipality or neighborhood association to validate workflows and data freshness SLAs. -*Responsive viewport design: Implementing mobile and differing viewport responsive UI/UX