Hybrid RAG Knowledge Engine
Combines multi-tenant pgvector semantic search with relational SQL filters, parsing developer brochures and payment plans so the AI answers accurately without hallucinations.
AI & AUTOMATION / REAL ESTATE
PlotLeadAI
From scattered leads to one connected sales journey.
01The challenge
A real problem
worth solving.
Real estate agencies lose high-intent leads between advertising, WhatsApp conversations, follow-ups, agents, and appointments. Generic LLMs hallucinate property prices, conflate developments, and lack agency inventory knowledge. The business required an enterprise revenue system connected to a domain-engineered RAG knowledge base, from lead capture to automated AI qualification, appointment booking, attribution, and sales visibility.
Our role
Founder and sole engineer: designed the multi-tenant SaaS architecture, engineered the hybrid RAG knowledge base engine with pgvector, built the document ingestion pipeline for developer brochures, developed the automated AI qualification layer, campaign attribution engine, WhatsApp messaging service, and Stripe subscription infrastructure.
02The solution
The details that
make the difference.
AI-Assisted Lead Qualification
Conversations grounded in agency-specific property inventory and real-time availability via hybrid RAG, rather than generic chatbot responses.
Document Ingestion Pipeline
Automated processing of developer PDF brochures, price lists, and floor plans with semantic chunking and metadata enrichment.
Full Revenue Journey
Connects lead capture, qualification, follow-up, appointment, attribution, and sales in one system instead of disconnected tools.
Multi-Tenant Data Isolation
Each agency's knowledge base, vector embeddings, leads, and properties are isolated at the tenant level with Supabase RLS.
Autonomous WhatsApp Infrastructure
Dedicated messaging service powered by Google Gemini and WhatsApp API handling conversational qualification and agent alerts.
Campaign Attribution & ROI
Tracks which campaigns, sources, and touchpoints drive actual closed sales, not just clicks or impressions.
Subscription Billing
Complete Stripe integration with tiered plan management, usage tracking, and automated billing lifecycle.
Explore the full delivery scope
- Multi-tenant CRM with real-time lead pipeline and stage management
- Hybrid RAG knowledge base engine with isolated pgvector similarity search per agency
- Document ingestion pipeline parsing PDF brochures, developer specs, and payment plans
- Semantic chunking with property metadata tagging (developer, price, location, delivery date)
- Dual hybrid search combining dense vector embeddings with structured relational SQL filters
- Anti-hallucination guardrails enforcing factual grounding in verified agency listings
- Cross-lingual Arabic and English semantic retrieval for MENA real estate inquiries
- AI-assisted qualification conversations grounded in live property database inventory
- Autonomous WhatsApp AI agent service with QR linking and asynchronous queue processing
- Property management system with dynamic public landing pages
- Multi-touch campaign attribution tracking cost-per-lead across marketing channels
- Automated email follow-up sequences and multi-step lead nurture workflows
- Team roles, invitation workflows, and granular permission management
- Stripe subscription billing infrastructure with tiered plan limits and customer portal
- Tenant-isolated data architecture with Supabase Row Level Security (RLS)
Under the hood
Hybrid RAG Knowledge Base Architecture
Generic AI chatbots fail in real estate because they hallucinate prices, conflate projects, and lack live agency inventory. PlotLeadAI deploys a domain-engineered Retrieval-Augmented Generation (RAG) engine that isolates knowledge per agency tenant and combines unstructured vector embeddings with structured relational database queries.
Document Ingestion & Parsing
Automated extraction pipeline for PDF developer brochures, unit specifications, floorplans, and payment milestone schedules.
Semantic Chunking & Metadata Tagging
Splits property documents into contextual semantic chunks enriched with relational tags: developer, price range, property type, location, and delivery date.
Multi-Tenant pgvector Isolation
Vector embeddings stored with tenant-level Row Level Security (RLS), guaranteeing complete isolation of proprietary agency knowledge.
Hybrid Search & Relational Filtering
Executes cosine similarity search concurrently with hard SQL filters (exact budget, bedroom count, handover year) to eliminate semantic drift.
Anti-Hallucination Guardrails
Enforces strict context boundaries: the AI qualifier only cites verified listing data and falls back gracefully when inventory does not match.
Bilingual MENA Retrieval
Cross-lingual Arabic and English embeddings supporting local real estate vernacular and dialect terms across Gulf and Levant markets.
03Inside the product
The actual experience.
13 product screens.
Select any image for a closer look.
The thinking behind the build
PlotLeadAI was engineered around the complete revenue journey rather than a collection of disconnected features. Its intelligence is powered by an isolated, multi-tenant RAG knowledge base engine grounded in real property documents and structured inventory data rather than generic chatbots.
Your business, next