Jorge Real Estate Bots

40% of real estate leads go cold because agents take >5 minutes to respond. Jorge uses one unified conversation system with specialized seller, buyer, and lead-intake handlers behind a single routing layer.

Production system — Automated lead qualification for a real estate agency using GHL webhooks, Claude AI, and Redis. 1,707 tests passing.

Built for Jorge Salas (Acuity Real Estate, Rancho Cucamonga CA) by Cayman Roden / Roden AI Solutions. Processed 500+ inbound leads via GoHighLevel CRM (Jan-Mar 2026) with zero downtime. This is not a tutorial or demo -- it was a deployed, revenue-generating AI system.

Production Dashboard (Lyrio)

Chat Interface	Lead Activity	Bot Management

Cost Tracking	Lead Pipeline	Tone Settings

What This Solves

• Missed leads -- The app responds within seconds, not minutes, and routes each contact into the correct qualification path
• Manual qualification is slow -- Structured seller and buyer flows extract budget, timeline, pre-approval status, motivation, condition, and price without agent involvement
• No pipeline visibility -- Admin and dashboard APIs expose canonical mode, status, handoff reason, and next-step state for every conversation

Key Metrics

Metric	Value
Tests	1,707 passing
Public model	1 canonical conversation system
Cross-Bot Handoff	0.7 confidence threshold, circular prevention, rate limiting
CRM Integration	GoHighLevel real-time sync
Temperature Scoring	Hot/Warm/Cold with automated tag publishing
AI Routing	Claude Haiku/Sonnet model selection
Deploy	Render (single unified app) + Docker Compose

Business Impact

Metric	Value
Leads qualified	500+ since January 2026
Uptime	Zero downtime since launch (24/7)
Response time	<500ms average
Languages	English and Spanish (no additional staffing)

Production Metrics

Verified operational data from production deployment (Jan-Mar 2026):

System	Metric	Value	How Verified
Lead Processing	Total leads handled	500+	GHL contact records
Uptime	Downtime incidents	0	Render health checks, 24/7
Response Latency	Average response time	<500ms	FastAPI request logging
Webhook Reliability	Deduplication	Two-phase TTL (120s guard + 300s post-success)	routes_webhook.py dedup keys
Webhook Reliability	Per-contact lock	Atomic setnx, 90s TTL	Prevents concurrent message handling
Rate Limiting	Global	Per-minute + per-endpoint	rate_limit_middleware.py
Rate Limiting	Per-contact	10 msgs/min	Redis counter with TTL
Model Routing	Haiku	Routine tasks (lead categorization)	claude_client.py TaskComplexity enum
Model Routing	Sonnet	Complex analysis (qualification)	Cost/quality-aware routing
Model Routing	Opus	High-stakes (seller negotiations)	Reserved for critical decisions
Prompt Caching	Anthropic cache	Enabled for >1024 char system prompts	cache_read_input_tokens tracking
Circuit Breaker	GHL API protection	Opens after 5 failures in 60s, 30s cooldown	GHLCircuitBreaker class
Conversation History	Context window	20 messages (10 turns) max	Prevents context bloat
Response Safety	Identity filters	38 regex patterns	response_filter.py
Response Safety	Output truncation	480 chars at word boundary	SMS-compatible responses
Bilingual	Spanish detection	2+ indicator words from frozen set	Auto-routes to BILINGUAL_HANDOFF
Test Coverage	Passing tests	1,707	pytest tests/ -q
Retry Logic	Anthropic API	Exponential backoff (2s-15s) via tenacity	RateLimitError + InternalServerError

For Hiring Managers

If you're evaluating for...	Where to look	Training behind it
AI / ML Engineer	Claude conversation engine (bots/shared/claude_client.py), confidence-based model routing (bots/shared/business_rules.py), multi-turn memory management	IBM GenAI Engineering (144h), Microsoft AI & ML Engineering (75h)
Backend / AI Automation Engineer	Webhook normalization + dedup + per-contact locking (bots/lead_bot/routes_webhook.py), Redis rate limiting, GHL CRM real-time sync	IBM GenAI Engineering (144h), Vanderbilt Agentic AI (58h)
CRM / Marketing Automation	Full AWARENESS→CONVERSION funnel attribution (bots/shared/funnel_attribution.py), SMS re-engagement sequences (bots/shared/stall_reengagement.py), campaign metrics (bots/shared/sms_metrics_collector.py)	Google Digital Marketing (190h), Meta Social Media Marketing (83h)
Data Analyst / BI	Lead dashboard API (bots/lead_bot/routes_dashboard.py), funnel conversion metrics, per-stage ROI tracking	Google Data Analytics (181h), IBM BI Analyst (141h)

→ Full cert-to-code mapping: docs/certifications.md (920h across 8 certifications)

API Overview

30+ endpoints across webhook routing, lead dashboard, admin controls, and real-time events:

Architecture

The system runs as a single FastAPI application. Incoming webhooks are normalized, deduplicated, locked per-contact, resolved to a canonical mode, and then dispatched to specialized internal handlers.

%%{init: {'theme': 'dark', 'themeVariables': {'primaryColor': '#D4A574', 'primaryBorderColor': '#C1440E', 'primaryTextColor': '#E2E8F0', 'lineColor': '#C1440E', 'background': '#1A1510'}}}%%
flowchart TB
  subgraph Incoming["Incoming Leads"]
    Web["Web Forms"]
    GHLHook["GHL Webhooks"]
    API["REST API"]
  end

  subgraph App["Unified App :8001"]
    Webhook["routes_webhook.py\nUnified orchestrator"]
    Dashboard["routes_dashboard.py\n12 dashboard endpoints"]
    Admin["routes_admin.py\nBot config & state"]
    Realtime["routes_realtime.py\nWebSocket events"]
    Lead["Lead Intake\nintent analysis"]
    Buyer["Buyer Handler\nQ0-Q4, property matching"]
    Seller["Seller Handler\nQ1-Q4, pricing"]
  end

  subgraph Intelligence["AI & Decision Engine"]
    Intent["Intent Decoder\nRegex + semantic analysis"]
    Temp["Temperature Scoring\nHot >=80 | Warm 40-79 | Cold <40"]
    Claude["Claude AI\nHaiku/Sonnet routing"]
    Funnel["Funnel Attribution\nRedis sorted sets, 30-day TTL"]
  end

  subgraph Infra["Infrastructure"]
    Redis[(Redis Cache)]
    GHL["GoHighLevel CRM\nTag publishing\nWorkflow triggers"]
  end

  subgraph Dashboard_UI["Monitoring"]
    Lyrio["Lyrio Dashboard\nStreamlit Cloud"]
  end

  Web --> Webhook
  GHLHook --> Webhook
  API --> Webhook

  Webhook --> Lead
  Webhook --> Buyer
  Webhook --> Seller

  Lead --> Intent
  Buyer --> Intent
  Seller --> Intent

  Intent --> Temp
  Intent --> Claude

  Lead --> Redis
  Buyer --> Redis
  Seller --> Redis
  Funnel --> Redis

  Temp --> GHL
  Lyrio --> Dashboard

Loading

Domain Context

Real estate agencies lose 40-60% of leads due to slow follow-up. This system:

• Responds to incoming GHL webhooks within seconds
• Qualifies leads using Claude AI conversation analysis
• Books appointments directly into the agency calendar
• All integrations configurable via environment variables — no code changes needed for new verticals

Quick Start

git clone https://github.com/ChunkyTortoise/jorge_real_estate_bots.git
cd jorge_real_estate_bots
pip install -r requirements.txt

# Start the unified app
uvicorn bots.lead_bot.main:app --host 0.0.0.0 --port 8001

Docker

cp .env.example .env
# Edit .env with your API keys

docker compose up
# App on :8001, Dashboard on :8501

Render Deployment

Production ran on Render (Jan-Mar 2026), processing 500+ leads with zero downtime. The repo includes render.yaml for Render Blueprint deployment. To redeploy, connect the repo and configure the jorge-env environment group with: REDIS_URL, GHL_API_KEY, ADMIN_API_KEY, ANTHROPIC_API_KEY, GHL_LOCATION_ID, JORGE_USER_ID, JORGE_CALENDAR_ID

Bot Capabilities

Lead Intake -- Semantic intent analysis plus conservative routing into seller, buyer, bilingual handoff, or human handoff.

Seller Handler -- Structured Q1-Q4 seller qualification, pricing and condition extraction, temperature scoring, and bounded handoff/escalation behavior.

Buyer Handler -- Full buyer qualification flow (Q0-Q4), preference extraction, temperature scoring, and weighted property matching.

Tech Stack

Layer	Technology
API	FastAPI (APIRouter pattern), Pydantic, uvicorn
Dashboard	Streamlit (Lyrio), Plotly
AI	Claude (Haiku/Sonnet routing)
Cache	Redis (sorted sets, rate limiting, bot state, funnel attribution)
CRM	GoHighLevel (webhooks, custom fields, workflows)
Testing	pytest, pytest-asyncio (1,707 tests passing)

Project Structure

Directory layout (click to expand)

jorge_real_estate_bots/
├── bots/
│   ├── shared/              # Config, Claude client, GHL client, cache, auth,
│   │                        # funnel_attribution, stall_reengagement,
│   │                        # bot_metrics_collector, alerting_service,
│   │                        # sms_metrics_collector, response_filter
│   ├── lead_bot/
│   │   ├── main.py          # FastAPI app, includes all routers
│   │   ├── routes_webhook.py     # GHL webhook dispatcher (unified + new-lead)
│   │   ├── routes_dashboard.py   # 12 dashboard/alert endpoints
│   │   ├── routes_admin.py       # Bot settings, reassign, reset state
│   │   ├── routes_realtime.py    # WebSocket events, recent events
│   │   ├── routes_productization.py  # Playbooks, reports
│   │   └── routes_test_endpoints.py  # Hardening test endpoints
│   ├── seller_bot/          # Q1-Q4 qualification, CMA engine
│   └── buyer_bot/           # Buyer qualification + property matching
├── database/                # SQLAlchemy models, async session
├── command_center/          # Streamlit dashboard components
├── tests/                   # 1,707 passing tests
├── docker-compose.yml       # Redis + app + dashboard
├── render.yaml              # Render Blueprint config
└── Dockerfile

API Endpoints

All endpoints are served from a single app on port 8001. Run uvicorn bots.lead_bot.main:app and visit /docs for the interactive reference.

Full endpoint listing (click to expand)

Webhooks (routes_webhook.py)

• POST /ghl/webhook/new-lead -- Compatibility entrypoint for new leads
• POST /api/ghl/webhook -- Unified inbound webhook and canonical routing/orchestration path
• POST /api/ghl/webhook/message-status -- SMS delivery status callbacks

Dashboard (routes_dashboard.py)

• GET /api/dashboard/metrics -- System + performance metrics
• GET /api/dashboard/leads/summary -- Hero metrics + conversation summary
• GET /api/dashboard/leads -- Paginated lead list (filterable by temperature)
• GET /api/dashboard/leads/{contact_id} -- Single lead detail
• GET /api/dashboard/handoffs -- Recent handoff events (with contact_id)
• GET /api/dashboard/conversations/{contact_id} -- Q&A transcript
• GET /api/dashboard/costs -- Cost/ROI data, commission pipeline
• GET /api/dashboard/sms-metrics -- SMS delivery stats (7-day rolling)
• GET /api/dashboard/funnel -- Funnel conversion by stage (AWARENESS through CONVERSION)
• GET /api/dashboard/stall-stats -- Stall re-engagement stats (optional ?contact_id filter)
• GET /api/alerts/active -- Active alerts
• POST /api/alerts/{alert_id}/acknowledge -- Acknowledge an alert

Admin (routes_admin.py)

• GET /admin/settings -- Current bot settings
• POST /admin/reassign-bot -- Reassign contact to a canonical mode
• PUT /admin/settings/{bot} -- Update bot configuration
• DELETE /admin/reset-state/{bot}/{contact_id} -- Reset conversation state
• GET /admin/conversations/{contact_id} -- Canonical conversation state detail

Real-time (routes_realtime.py)

• GET /api/events/recent -- Recent events (filterable, since_minutes, event_types)
• GET /api/events/ws-status -- WebSocket connection health
• GET /api/events/health -- Event system health (Redis check)
• WS /ws/events -- Live event stream

curl Examples

Analyze a new lead:

curl -X POST http://localhost:8001/analyze-lead \
  -H "Content-Type: application/json" \
  -d '{
    "contact_id": "abc123",
    "name": "Maria Santos",
    "email": "maria@example.com",
    "phone": "+1-555-0142",
    "message": "Looking to buy a 3BR home in Coral Gables under $650k. Pre-approved with Chase.",
    "source": "website"
  }'

Get funnel conversion data:

curl http://localhost:8001/api/dashboard/funnel \
  -H "X-Admin-Key: YOUR_ADMIN_KEY"

Get stall stats:

curl http://localhost:8001/api/dashboard/stall-stats \
  -H "X-Admin-Key: YOUR_ADMIN_KEY"

Troubleshooting

See docs/TROUBLESHOOTING.md for solutions to common issues: GHL webhook setup, Redis connection errors, environment variable checklist, HTTP error codes, and bot handoff failures.

Production Handoff Docs

Production Handoff Documentation (for operators -- click to expand)

The canonical production-finalization and handoff package lives in docs/:

• JORGE_V2_PRODUCTION_HARDENING_SPEC.md
• JORGE_OPERATOR_RUNBOOK.md
• GHL_CONFIGURATION_CONTRACT.md
• MIGRATION_CHECKLIST_CANONICAL_STATE.md
• COMPATIBILITY_SHIMS.md
• JORGE_GHL_WORKFLOW_INVENTORY.md
• JORGE_GHL_EXPORT_CAPTURE.md
• JORGE_LIVE_VALIDATION_CHECKLIST.md
• JORGE_PRODUCTION_HANDOFF_SIGNOFF.md

Helper scripts for finish-line verification:

# Verify canonical DB schema against a live database
DATABASE_URL=... python scripts/check_conversation_schema.py

# Hit deployed health/admin/dashboard endpoints and emit a readiness report
ADMIN_API_KEY=... JORGE_CONTACT_ID=... python scripts/production_readiness_report.py --base-url https://jorge-realty-ai-xxdf.onrender.com

# Validate live or exported GHL tags/custom fields/workflows against the contract
GHL_API_KEY=... GHL_LOCATION_ID=... python scripts/validate_ghl_contract.py --ghl-api-key "$GHL_API_KEY" --location-id "$GHL_LOCATION_ID"

# Export the live workflow list and heuristic workflow-risk summary
GHL_API_KEY=... GHL_LOCATION_ID=... python scripts/export_ghl_workflows.py --ghl-api-key "$GHL_API_KEY" --location-id "$GHL_LOCATION_ID" --json-output docs/ghl_workflows_export.json --md-output docs/ghl_workflows_export.md

# Generate a dry-run plan for creating missing canonical GHL tags/fields
GHL_API_KEY=... GHL_LOCATION_ID=... python scripts/sync_ghl_contract.py --ghl-api-key "$GHL_API_KEY" --location-id "$GHL_LOCATION_ID"

# Review extra live GHL tags/fields for likely routing or handoff risk
GHL_API_KEY=... GHL_LOCATION_ID=... python scripts/review_ghl_legacy_contract.py --ghl-api-key "$GHL_API_KEY" --location-id "$GHL_LOCATION_ID" --output docs/ghl_legacy_contract_review.md

# Run the repo-side production-finalization helpers and write outputs to docs/
JORGE_LIVE_URL=... ADMIN_API_KEY=... JORGE_CONTACT_ID=... DATABASE_URL=... bash scripts/run_production_finalization.sh

Testing

pytest tests/ -v                    # Full suite
pytest tests/shared/ -v             # Shared services
pytest tests/api/ -v                # Dashboard & API routes
pytest tests/lead_bot/ -v           # Realtime events

Changelog

See CHANGELOG.md for release history.

Related Projects

• EnterpriseHub -- Full real estate AI platform this was extracted from, with BI dashboards and CRM integration
• Lyrio Dashboard -- Streamlit analytics dashboard with AI concierge, connected to Jorge API
• ai-orchestrator -- AgentForge: unified async LLM interface (Claude, Gemini, OpenAI, Perplexity)

Built By

Developed by Cayman Roden (Roden AI Solutions) as a production engagement for Jorge Salas (Acuity Real Estate). The system has been in continuous production since January 2026.

• GitHub: ChunkyTortoise
• Portfolio: EnterpriseHub

License

MIT