Devpost: https://devpost.com/software/soft-submission

RentScope is a map-first assistant that helps students choose neighbourhoods by visualizing rent, commute, safety signals, and amenities; producing ranked, explainable recommendations instead of scattered listings.

Finding student housing is chaotic:

• Rent alone doesn’t tell the full story
• Commute time, safety, and amenities matter differently for everyone
• “Safety” and “livability” are hard to compare objectively
• Most tools force tab-hopping and gut-feel decisions

RentScope turns those tradeoffs into a calm, visual map experience with clear reasoning behind every recommendation.

• Students add the places they care about most (campus, work, etc.)
• RentScope generates:
  • A rent heatmap
  • Ranked neighbourhood shortlists
  • Evidence-backed explanations for each score
• Users can explore neighbourhoods, compare tradeoffs, and understand why one area outranks another

• frontend/ — Next.js + Tailwind + Leaflet
• orchestrator/ — Node.js API + Gemini agent manager
• mcp_server/ — Python FastAPI MCP server + tools
• data/ — sample datasets + Snowflake SQL scripts

• Frontend -> Orchestrator only
• Orchestrator -> MCP tools only (no Snowflake direct)
• Agents only interpret MCP tool outputs (JSON-only, validated)
• MongoDB stores shortlists, tool cache, and analysis runs

• Toronto Police Service Open Data (Major Crime Indicators / Public Safety Data Portal):
  • https://www.tps.ca/data-maps/open-data/
  • ArcGIS Feature Service (Major Crime Indicators open data): https://services.arcgis.com/S9th0jAJ7bqgIRjw/arcgis/rest/services/Major_Crime_Indicators_Open_Data/FeatureServer/0
• TTC GTFS (routes/stops/schedules):
  • https://data.urbandatacentre.ca/catalogue/city-toronto-ttc-routes-and-schedules
• OpenStreetMap / Overpass API (optional for POIs):
  • https://wiki.openstreetmap.org/wiki/Overpass_API

1. Create an environment variable with your Gemini key (PowerShell):

$env:GEMINI_API_KEY="your_key_here"

2. Run the stack:

docker compose up --build

3. Open the app: http://localhost:3000

cd mcp_server
python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt
uvicorn app:app --reload --port 7000

cd orchestrator
npm install
$env:MCP_URL="http://localhost:7000"
$env:MONGO_URI="mongodb://localhost:27017"
$env:GEMINI_API_KEY="your_key_here"
npm run dev

cd frontend
npm install
$env:NEXT_PUBLIC_ORCHESTRATOR_URL="http://localhost:4000"
npm run dev

Use the built-in demo shortlist button in the UI, or call the API directly:

curl -X POST http://localhost:4000/api/analyze \
  -H "Content-Type: application/json" \
  -d @data/sample/seed_demo.json

Optional: seed MongoDB with the same demo shortlist:

pip install pymongo
python data/seed_mongo.py

• crime_summary(lat, lon, radius_m, window_days)
  • returns {counts_by_type, rate_hint, trend_hint, source, updated_at}
• commute_proxy(lat, lon, campus_lat, campus_lon)
  • returns {distance_km, est_minutes, near_transit_hint, source}
• nearby_pois(lat, lon, categories, radius_m)
  • returns {results:[{name, category, dist_m}], counts_by_category, source}

SQL scripts live in data/snowflake/:

• create_tables.sql
• load_sample.sql

Set these env vars for Snowflake access (MCP server):

SNOWFLAKE_ACCOUNT
SNOWFLAKE_USER
SNOWFLAKE_PASSWORD
SNOWFLAKE_WAREHOUSE
SNOWFLAKE_DATABASE
SNOWFLAKE_SCHEMA
SNOWFLAKE_ROLE (optional)

• shortlists
• analysis_cache
• analysis_runs

• This MVP reports “higher/lower incident density” based on available open data only.
• Crime locations in the demo data are approximate; treat results as indicative.
• Agent outputs are validated JSON with retries; deterministic fallbacks run if Gemini is unavailable.