Bringing AI into real estate operations.

India Informatics developed an AI platform addressing three areas of day-to-day work: speaking with customers, processing service orders and estimating property values.

Each area called for a different approach. Customer conversations needed a voice interface that could follow context. Incoming forms needed information extracted and checked before it entered business systems. Property valuations needed a model that could work with structured property and market data.

The project combined voice AI, document processing, retrieval-augmented generation (RAG) and machine learning around those needs.

The voice component supported outbound conversations for inspection scheduling, customer communication and lead follow-up. Scripts could take account of the property type, location and lead stage, while conversational memory helped maintain context across multiple exchanges.

Twilio handled voice calls. Whisper converted speech to text, GPT-4o and LangChain supported the conversation, and ElevenLabs generated spoken responses. The documented scope also included Google Calendar, call recording and logging, and a custom REST API for sending summaries and tags to the CRM.

These components were intended to help property teams manage routine conversations and retain useful records of each interaction.

The document module was designed to identify the form type, extract its fields and populate the internal order-management system. The scope covered inspection, appraisal and maintenance forms arriving as PDFs through email or Google Drive.

LayoutLMv3 and Tesseract supported document processing, with confidence scores and human review used to check extracted information. A Node.js API connected the workflow to the order-management system.

The documented stack also included FAISS and GPT-4o for retrieval-augmented responses. Retrieval and field extraction are described as distinct parts of the technical scope; no extraction-accuracy claim is made here.

The valuation component used property characteristics and market information to produce an estimated value. Inputs included lot size, bedroom and bathroom counts, condition, property type, location, market trends and recent sales.

An XGBoost regression model, supported by a custom preprocessing pipeline, formed the core of the system. The documented functionality included API access, confidence intervals, suggestions for comparable properties and downloadable PDF valuation reports.

Data-source scope: The project notes list the internal CRM, Zillow API and county records, together with an MLS mock integration. The MLS connection was a mock, not a production feed. Use of the listed external sources in the Qatar deployment remains unconfirmed.