Back to work

AI Real Estate Investment Platform

Intelligent property discovery and portfolio analytics with conversational AI

Role

End-to-end (Frontend, Backend, Data Pipeline, Infrastructure)

Problem

Real estate investors needed a faster way to discover properties, analyze markets, and make data-driven decisions without manually searching through fragmented data sources.

This platform reimagines how real estate investors discover and analyze properties by combining conversational AI with rich market data. The system enables natural-language queries like 'Show me undervalued properties in growing neighborhoods' and returns actionable insights backed by demographics, market trends, and custom scoring models. Built as a full-stack SaaS with multitenancy, subscription billing, and KYC flows, it handles everything from address normalization to fraud detection at scale.

The Challenge

Traditional real estate platforms require users to filter through dozens of fields, manually cross-reference market data, and jump between tools to get a complete picture. Investors waste time on repetitive research tasks instead of evaluating opportunities. The goal was to build an AI-first platform that could understand investor intent, fetch relevant data in real-time, and present insights conversationally while maintaining accuracy and citation.

Technical Architecture

Frontend

  • Next.js 15 (App Router)
  • React 19
  • TypeScript
  • TailwindCSS v4
  • Framer Motion
  • Mapbox GL

Backend & APIs

  • Python (FastAPI)
  • OpenAI (GPT-4 + Embeddings)
  • LangChain / LangGraph

Data & Search

  • PostgreSQL (Supabase)
  • Elasticsearch
  • pg-boss (Job Queue)
  • Redis (Rate Limiting)

Infrastructure

  • Vercel (Frontend + Cron)
  • Railway (Microservice)
  • Fly.io (WebSocket)
  • Docker

Integrations

  • Stripe (Subscriptions)
  • Plaid (KYC/IDV)
  • Knock (Notifications)
  • Resend (Email)
  • Clerk (Auth)

Key Features Built

  • Conversational AI chat with streaming responses, function calling, and contextual memory
  • Property analysis tools: market trends, offer analysis, maintenance projections, commercial viability
  • Watchlist system for tracking properties with bulk CSV import and enrichment
  • Address resolution microservice: Elasticsearch-powered fuzzy matching with normalization
  • Background job orchestration: pg-boss for async batch processing with retry logic
  • Interactive maps: Clustering, heatmaps, region boundaries, and demographics overlays
  • Portfolio snapshots: Save and compare investment performance over time
  • Subscription billing: Stripe integration with tiered data bundles and usage metering
  • KYC flow: Plaid IDV integration for investor verification
  • Org-scoped onboarding with progress tracking and automated email reminders (Vercel Cron)

Technical Challenges & Solutions

Challenge

Address normalization with high variance (abbreviations, typos, missing fields).

Solution

Built multi-pass Elasticsearch query with progressive fuzzy matching and confidence scoring.

Challenge

Slow batch processing for 100K+ addresses.

Solution

Implemented parallel job workers with configurable concurrency and backpressure tuning (pg-boss).

Challenge

AI hallucinations and unsourced claims.

Solution

Constrained model to only use function-calling tools, return cited data points, and surface confidence scores.

Challenge

Map performance with 10K+ markers.

Solution

Implemented supercluster for client-side clustering and lazy-loaded region data via viewport bounds.

←Civic Outreach & Donor Operations PlatformAI Document Review & Q&A Portal→