ADREI (AI-Driven Real Estate Intelligence) is an advanced property analysis platform that uses a sophisticated multi-agent AI framework with Bright Data APIs for web scraping, Google Gemini AI for intelligent data extraction and analysis, and Mapbox for location-based filtering.

• Intelligent Query Caching:
  • AI detects similar queries (e.g., "apartments near UC Berkeley" = "housing near Berkeley campus")
  • Automatically reuses previously scraped data
  • Skips Stage 1 & 2 entirely for duplicate searches
  • Saves API costs and time!
• Four-Stage Processing:
  • Stage 0: Check query cache with AI similarity detection
  • Stage 1: Quick scan for address, price, and URL (if not cached)
  • Stage 2: Parallel deep scrape - Sends all requests to Bright Data simultaneously (6-8x faster!) + AI Detail Extraction (bedrooms, bathrooms, amenities, utilities, contact info) stored in metadata
  • Stage 3: Distance-based filtering with AI-powered location understanding + Mapbox geocoding
  • Stage 4: AI-powered property management quality analysis & scoring (0-100)
• Intelligent Search: Uses Bright Data's SERP API to search Google for apartment listings
• Multi-Site Scraping: Automatically visits and scrapes multiple apartment listing websites
• AI-Powered Parsing: Leverages Google Gemini to extract structured data from diverse HTML layouts
• Location Intelligence:
  • AI corrects spelling mistakes and typos automatically
  • Expands abbreviations (e.g., "MIT" → full name with location)
  • Calculate walking distances from any location (universities, workplaces, etc.)
  • Fallback to straight-line distance if route calculation fails
• Smart Caching:
  • Avoids re-scraping already visited apartments
  • Caches geocoded coordinates for all apartments
  • Caches distance calculations between apartment + target location pairs
  • Caches management analysis results permanently
  • Up to 90% API cost savings on repeat queries!
• Unique IDs: Each apartment gets a deterministic ID for linking data from multiple sources
• Price Filtering: Filter by price (e.g., "under $2000")
• Rich Data Storage: Saves complete HTML and comprehensive metadata for future reference
  • Basic info: address, price, URL
  • Apartment details: bedrooms, bathrooms, square feet, deposit, lease length
  • Amenities: pool, gym, parking, etc.
  • Utilities: water, electricity, internet (included or tenant pays)
  • Contact info: phone, email, office hours
  • Pet policy and parking details
• Management Analysis:
  • AI-powered reputation research
  • Quality scoring (0-100 scale)
  • Strengths, concerns, and recommendations
  • Permanently cached to avoid re-analysis
• 🆕 Detailed Apartment Viewer:
  • Query "look into apartment 3" to see ALL details for a specific apartment
  • Shows bedrooms, bathrooms, amenities, utilities, contact info
  • Displays comprehensive scores from all research agents
  • Shows management analysis with key strengths and concerns
• 🆕 4 Specialized Research Agents (Micro-Location Analysis):
  • Security Agent: Street lighting, crime, foot traffic, security features (5 sub-metrics)
  • Accessibility Agent: Sidewalks, healthcare, transit, wheelchair access (5 sub-metrics)
  • Pet Friendliness Agent: Parks, vets, pet stores, walking quality (5 sub-metrics)
  • Lifestyle Agent: Dining, nightlife, entertainment, shopping, fitness (5 sub-metrics)
  • Each agent analyzes specific street + 2-3 block radius (not city-wide!)
  • Provides overall score + multiple sub-metrics + key findings + detailed reasoning
  • Answers ANY question: "is it safe?", "good for pets?", "wheelchair accessible?", "good nightlife?"
  • Cached permanently (instant results on re-query)
• High Performance:
  • Parallel processing in Stage 2 (all apartments scraped simultaneously)
  • 6-8x faster than sequential scraping
  • Bright Data cloud handles concurrent requests efficiently

                        STAGE 0: INTELLIGENT QUERY CACHE
User Query → Gemini AI (Normalize Query) → Check Query Cache → 
Compare with Previous Queries → [MATCH?] → Load Cached Apartments
                                    ↓ [NO MATCH]
                                    
                        STAGE 1: DISCOVERY
SERP API (Google Search) → Filter Relevant Sites → 
Web Unlocker (Scrape) → Gemini AI (Extract Address/Price/URL) → 
Price Filter

                        STAGE 2: PARALLEL COLLECTION
Batch All Apartments → Parallel Requests to Bright Data Cloud → 
Wait for All Responses → Save Raw Markdown + Metadata → Generate Unique IDs →
Save Query to Cache

                        STAGE 3: LOCATION FILTERING (Optional)
User Distance Query → Gemini AI (Interpret & Correct Location) → 
Mapbox Geocoding (Target Location) → Mapbox Geocoding (Each Apartment) → 
Calculate Walking Distances → Filter by Radius → Save Distance Data

                        STAGE 4: MANAGEMENT ANALYSIS (Optional)
User Details Query → Load Apartment Markdown → Gemini AI (Extract Management Name) →
Gemini AI (Analyze Reputation & Quality) → Score 0-100 → 
Save Analysis to Metadata → Display Results

1. Bright Data Account (https://brightdata.com)

   • API Key
   • SERP Zone configured
   • Web Unlocker Zone configured

2. Google AI Studio Account (https://ai.google.dev/)

   • Google API Key for Gemini

3. Mapbox Account (https://www.mapbox.com/)

   • Mapbox Access Token (for geocoding and distance calculations)

4. Node.js (v16 or higher)

cd "C:\Studies\Project\CAL HACKS 12.0\Room Hunt"

npm install

This will install:

• dotenv - Environment variable management
• node-fetch - HTTP client for API calls
• @google/generative-ai - Google Gemini AI SDK
• @mapbox/mapbox-sdk - Mapbox Geocoding & Directions API
• readline - CLI interface

Create a .env file in the project root with the following:

# Bright Data API Configuration
BRIGHTDATA_API_KEY="your_brightdata_api_key_here"
BRIGHTDATA_SERP_ZONE="your_serp_zone_id_here"
BRIGHTDATA_UNLOCKER_ZONE_ID="your_unlocker_zone_id_here"

# Google AI Configuration
GOOGLE_API_KEY="your_google_api_key_here"

# Mapbox Configuration (Optional - hardcoded token included, but you can override)
MAPBOX_ACCESS_TOKEN="your_mapbox_access_token_here"

Bright Data:

1. Log in to your Bright Data dashboard
2. Go to Zones → Find your SERP Zone → Copy the Zone ID
3. Go to Zones → Find your Web Unlocker Zone → Copy the Zone ID
4. Go to Settings → API → Copy your API Key

Google Gemini:

1. Visit https://ai.google.dev/
2. Click Get API Key
3. Create a new project or select existing
4. Copy your API key

Mapbox:

1. Visit https://www.mapbox.com/
2. Sign up for a free account
3. Go to Account → Access tokens
4. Copy your default public token (or create a new one)
5. Note: Free tier includes 100,000 requests/month

Test Bright Data connectivity:

npm test

This runs the demo script to verify your API credentials work.

Start the ADREI web server:

npm run web

Then open your browser to: http://localhost:3000

Server file: web-server.js

Features:

• 🎨 Professional white-themed UI with intelligent chat assistant
• 🏠 Dynamic apartment cards with comprehensive property data
• 💬 Interactive ADREI Assistant (same intelligence as CLI)
• 📊 Real-time updates via Server-Sent Events
• 🔍 Detailed property viewer with full analysis
• 🔗 Direct links to original listings

npm start

or

node apartment-finder.js

The application understands two types of queries:

Enter your query: apartments for rent near Chico State

Enter your query: 2 bedroom apartments in San Francisco under $3000

Enter your query: studio apartments near UCLA under $2000

Enter your query: affordable housing in Seattle

Enter your query: within 1 mile of UC Berkeley

Enter your query: within 1 mile of UC Berkley
(AI auto-corrects: "Berkley" → "Berkeley")

Enter your query: less than 2 miles from chico state
(AI expands: → "California State University Chico, Chico, California")

Enter your query: within 0.5 miles of Stanford

Enter your query: under 3 kilometers from MIT
(AI expands: → "Massachusetts Institute of Technology, Cambridge, Massachusetts")

✨ The AI handles typos and abbreviations automatically!

Enter your query: give me details

Enter your query: show me a report

Enter your query: provide information

Enter your query: analyze management

Enter your query: show reviews

Enter your query: quality score

✨ AI analyzes property management reputation and scores 0-100!

Enter your query: look into apartment 3

Enter your query: show me apartment 1

Enter your query: details for apartment 5

Enter your query: apartment 2 details

✨ Shows ALL info: beds/baths, amenities, utilities, contact, management analysis, research scores!

Enter your query: is it safe?
→ Security Agent researches crime, police presence, street safety for EACH apartment

Enter your query: good for pets?
→ Pet Friendliness Agent researches dog parks, vets, pet stores nearby

Enter your query: wheelchair accessible?
→ Accessibility Agent researches sidewalks, transit, healthcare access

Enter your query: good nightlife?
→ Lifestyle Agent researches dining, entertainment, nightlife

Enter your query: are there homeless in this area?
→ Security Agent analyzes homelessness issues for each specific street

✨ Each agent scores 0-100 + provides detailed findings for EACH apartment's street!

1. Query Normalization: Gemini AI extracts search intent (location, price, bedrooms, keywords)
2. Similarity Detection: Compares with previous queries to find matches
   • "apartments near UC Berkeley" matches "housing near Berkeley campus"
   • "rentals in Berkeley under $2000" matches "apartments in Berkeley under $2500"
3. Cache Hit: If similar query found, loads apartments instantly (skips Stage 1 & 2!)
4. Cache Miss: Proceeds to Stage 1 & 2

1. Search Phase: Searches Google for relevant apartment listing websites
2. Filtering Phase: Filters results to apartment-focused sites (excludes Zillow, social media, etc.)
3. Quick Scan: Scrapes top sites and extracts basic info:
   • Address
   • Price
   • Listing URL
4. Price Filtering: Filters by price if specified (e.g., "under $2000")
5. Parallel Deep Scrape:
   • Checks cache for previously scraped apartments
   • Batches all uncached apartments
   • Sends parallel requests to Bright Data (all at once!)
   • Waits for all responses simultaneously
   • Typically 6-8x faster than sequential scraping
6. Data Storage: Saves raw HTML and metadata with unique apartment ID
7. Query Caching: Saves this query's intent and results for future similar searches

1. Parse Query: Extracts distance (e.g., "1 mile") and target location (e.g., "UC Berkley")
2. AI Interpretation: Gemini AI corrects typos and clarifies location (e.g., "UC Berkley" → "University of California Berkeley, Berkeley, California")
3. Geocode Target: Converts corrected location to GPS coordinates using Mapbox
4. Geocode Apartments: Converts each apartment address to coordinates
5. Calculate Distances: Uses Mapbox Directions API to calculate walking distance (falls back to straight-line if needed)
6. Filter Results: Shows only apartments within specified radius
7. Save Distance Data: Stores distance info with same apartment ID for data linking
8. Display: Shows filtered results with distance and walking time

1. Detect Query: Recognizes keywords like "details", "report", "summary", "analysis", "reviews", etc.
2. Load Apartments: Loads current filtered/scraped apartment list from apartments_index.json
3. Check Cache: For each apartment, checks if management analysis already exists in metadata
4. Extract Management Name: Uses Gemini AI to identify property management company from markdown
5. AI Analysis: Gemini AI researches and evaluates management on multiple factors:
   • Reputation & trustworthiness
   • Responsiveness to tenant issues
   • Property maintenance quality
   • Lease terms fairness
   • Communication quality
   • Overall tenant satisfaction
6. Generate Score: AI assigns quality score (0-100) with rating (Excellent/Good/Fair/Below Average/Poor)
7. Create Report: Generates summary, strengths, concerns, and recommendation
8. Save & Cache: Permanently stores analysis in apartment metadata (never re-analyzed)
9. Display: Shows all apartments with management scores and reports

================================================================================
FINAL RESULTS
================================================================================

📁 All data saved to: ./apartment_data/

Each apartment has:
  - Raw Markdown file (.md) - cleaner than HTML!
  - Metadata file (.json) with address, price, URL, analysis

Summary of apartments:

1. [NEW] 123 Main St, Berkeley, CA 94704
   🆔 ID: apt_8f3a9c2b
   💰 Price: $2,500/month
   🔗 URL: https://...

2. [CACHED] 456 Oak Ave, Berkeley, CA 94705
   🆔 ID: apt_1d4e5f6a
   💰 Price: $2,800/month
   🔗 URL: https://...

================================================================================
STAGE 3: DISTANCE FILTERING
================================================================================

[Filter] Distance: 1 mile
[Filter] Location: UC Berkeley

[1/10] 123 Main St, Berkeley, CA 94704
  📏 Distance: 0.45 miles
  ⏱️  Walking time: 9 min walk
  ✅ Within range!

[2/10] 456 Oak Ave, Berkeley, CA 94705
  📏 Distance: 1.2 miles
  ⏱️  Walking time: 24 min walk
  ✗ Out of range (1.2 miles > 1 mile)

================================================================================
FINAL RESULTS
================================================================================

Summary of apartments:

1. [NEW] 123 Main St, Berkeley, CA 94704
   🆔 ID: apt_8f3a9c2b
   💰 Price: $2,500/month
   📏 Distance: 0.45 miles
   ⏱️  Walking: 9 min walk
   🔗 URL: https://...

================================================================================
STAGE 4: AI-POWERED MANAGEMENT ANALYSIS
================================================================================

[AI Agent] Analyzing 2 property managements...
This may take a few moments...

================================================================================
[1/2] 123 Main St, Berkeley, CA 94704
Price: $2,500/month
   Management: Berkeley Property Management LLC

[AI Agent] Analyzing management: "Berkeley Property Management LLC"
[AI Agent] ✓ Score: 82/100 (Good)
[Storage] ✅ Saved management analysis (Score: 82/100)

================================================================================
[Stage 4 Summary]
  ✓ Total properties: 2
  💾 From cache: 0
  🆕 Newly analyzed: 2
  ⚠️  Skipped: 0

================================================================================
FINAL RESULTS
================================================================================

Summary of apartments:

1. [NEW] 123 Main St, Berkeley, CA 94704
   🆔 ID: apt_8f3a9c2b
   💰 Price: $2,500/month
   📏 Distance: 0.45 miles
   ⏱️  Walking: 9 min walk
   🏢 Management: Berkeley Property Management LLC
   📊 Quality Score: 82/100 (Good)
   📝 Well-established company with responsive maintenance team. Generally positive 
       tenant reviews with some concerns about lease renewal processes.
      ✅ Strengths: Quick maintenance response, Professional staff
      ⚠️  Concerns: Lease renewal fees, Limited parking
   🔗 URL: https://...

Room Hunt/
├── apartment-finder.js        # Main orchestration & CLI 
├── api-services.js            # API integrations: Bright Data, Gemini, Mapbox 🆕
├── cache-manager.js           # Data storage & caching logic 🆕
├── brightdata_api_demo.js     # Reference implementation
├── package.json               # Node.js dependencies
├── .env                       # API keys (YOU CREATE THIS)
├── README.md                  # This file
├── MODULE_STRUCTURE.md        # Module organization guide 🆕
├── SETUP_CHECKLIST.md         # Setup guide
├── PROJECT_SUMMARY.md         # Technical documentation
├── STAGE3_DISTANCE_FILTERING.md  # Distance filtering guide
├── STAGE4_MANAGEMENT_ANALYSIS.md # Management analysis guide 🆕
├── INTELLIGENT_QUERY_CACHING.md  # Query caching details
└── apartment_data/            # Generated data directory
    ├── query_cache.json       # Cache of previous searches
    ├── apartments_index.json  # Master registry of all apartments
    ├── apt_8f3a9c2b.md        # Raw Markdown for apartment 🆕
    ├── apt_8f3a9c2b.json      # Metadata (address, price, URL, management_analysis) 🆕
    ├── apt_8f3a9c2b__distance.json  # Distance data (if filtered)
    ├── apt_1d4e5f6a.md
    ├── apt_1d4e5f6a.json
    └── ...

The codebase has been split into 3 focused modules for better maintainability:

1. api-services.js (440 lines)

   • All external API integrations (Bright Data, Gemini AI, Mapbox)
   • Exports: CONFIG, model, API functions

2. cache-manager.js (265 lines)

   • Data storage, caching, and file management
   • Handles query cache, URL cache, and file operations

3. apartment-finder.js (653 lines)

   • Main orchestration logic and CLI interface
   • Imports from both modules above

Benefits:

• ✅ Clear separation of concerns
• ✅ Easy to test individual components
• ✅ Better maintainability (vs. 1358-line monolith)
• ✅ Faster to locate specific functionality

See MODULE_STRUCTURE.md for detailed documentation!

• query_cache.json: Stores previous search queries with normalized intents and apartment IDs (enables intelligent query reuse)
• apartments_index.json: Master list of all scraped apartments with their unique IDs
• apt_XXXXXXXX.md: Complete raw Markdown from the apartment listing page (cleaner than HTML!)
• apt_XXXXXXXX.json: Extracted metadata including:
  • Basic info: address, price, URL, scrape timestamp
  • Distance data: coordinates, walking distance/time (if Stage 3 used)
  • Management analysis: company name, score, rating, summary, strengths, concerns (if Stage 4 used)
• apt_XXXXXXXX__distance.json: (Legacy) Distance calculations (now stored in main JSON)

In apartment-finder.js, modify the searchGoogle() call:

const serpResults = await searchGoogle(enhancedQuery, 10);  // Change 10 to desired number

In apartment-finder.js, adjust the slice:

const topResults = filteredResults.slice(0, 5);  // Change 5 to desired number

In apartment-finder.js, modify parseHtmlWithAI():

const maxLength = 30000;  // Change to adjust token usage

Solution: Ensure your .env file exists and has correct variable names

Solution:

• Verify your SERP Zone is active in Bright Data dashboard
• Check your API key has sufficient credits
• Ensure BRIGHTDATA_SERP_ZONE is correctly set

Solution:

• Verify your Web Unlocker Zone is active
• Check you have credits available

Solution:

• Verify GOOGLE_API_KEY is correct
• Ensure Generative Language API is enabled
• Check you haven't exceeded API quota

Solution:

• Some sites are heavily JavaScript-rendered
• Try different search queries
• The site structure may not have clear listing information

Old Sequential Approach:

• 10 apartments × (3-5 sec each + 3 sec delay) = 60-80 seconds
• 50 apartments = 5+ minutes 😴

New Parallel Approach:

• 10 apartments scraped simultaneously = ~5-10 seconds ⚡
• 50 apartments scraped simultaneously = ~10-15 seconds 🚀
• 6-8x faster overall!

How it works:

1. Filter out cached apartments (instant)
2. Send all uncached apartments to Bright Data at once
3. Bright Data cloud processes them in parallel
4. Save all results when complete

• Cost Awareness:
  • Each SERP request and Web Unlocker request consumes Bright Data credits
  • Parallel processing doesn't increase cost - same # of requests, just faster!
  • Mapbox Geocoding API: Free tier includes 100,000 requests/month
  • Google Gemini: Generous free tier for testing
• Rate Limits: Be mindful of API rate limits on Bright Data, Google AI, and Mapbox
• HTML Complexity: Very JavaScript-heavy sites might not return full content in initial HTML
• Token Limits: Large HTML files are truncated before sending to Gemini (100,000 chars limit)
• Caching: Already-scraped apartments are skipped to save API costs
• Unique IDs: SHA256-based deterministic IDs allow linking data from multiple sources
• Distance Calculations: Uses walking routes via Mapbox Directions API for real-world distances
• Ethical Scraping: Respect website terms of service and robots.txt

• Add caching to avoid re-scraping same URLs ✅ Implemented
• Add filtering options (price range) ✅ Implemented
• Add support for location-based radius searches ✅ Implemented with Mapbox
• Generate unique IDs for data linking ✅ Implemented
• Export results to CSV/Excel file
• Implement pagination support to get more listings from each site
• Add support for Bright Data Data Collector API for specific sites
• Implement parallel scraping for faster execution ✅ Implemented
• Add more filters (bedrooms, bathrooms, amenities)
• Add email/SMS alerts for new matching listings
• Add transit time calculations (bus, subway, etc.)

ADREI (AI-Driven Real Estate Intelligence System) was developed for CAL HACKS 12.0.
This project is for educational and demonstration purposes.

• Bright Data for providing powerful web scraping infrastructure
• Google for Gemini AI API
• Mapbox for geocoding and directions APIs
• Node.js community for excellent async patterns