AI Property Assistant

Per-listing chat assistant with two specialized agents and server-side memory.

Two Agents

Agent	Handles	Context
LISTING	Amenities, pitch advice, objection handling, comparisons	Property details, tags, location, sales intelligence
PRICING	Price justification, offer explanation, value comparison, cancellation plans	All of LISTING + full price breakdown, bank offers, nightly rates

Routing

The router picks the agent based on keywords:

"How should I pitch this to a family?"     → LISTING agent
"Why is it priced at ₹16,400?"             → PRICING agent
"What if they say it's too expensive?"      → PRICING agent (has "expensive")
"What amenities does it have?"              → LISTING agent
"What cancellation options are available?"  → PRICING agent (has "cancellation")

Server-Side Memory

Conversations are stored in PostgreSQL via Spring AI's JdbcChatMemoryRepository. Frontend just sends conversationId + message — no history to maintain.

POST /api/v1/crs/listings/assistant/{listingId}/chat
{
  "conversationId": "conv_abc123",
  "message": "How should I pitch this to a family?"
}

Follow-up with same conversationId:

{
  "conversationId": "conv_abc123",
  "message": "What if they say it's too expensive?"
}

The assistant remembers the previous question about families and connects it to the pricing response.

Response

{
  "listingId": "lst_abc",
  "message": "Start with the private pool and game room...",
  "agent": "LISTING"
}

agent field tells the frontend which agent answered.

Quick Suggestions

GET /assistant/{listingId}/start returns contextual suggestions based on listing data:

{
  "suggestions": [
    "Why is it priced at ₹16,400?",
    "How should I pitch this to a family?",
    "What objections might a client raise?",
    "How does this compare to similar villas?",
    "What cancellation options are available?",
    "What bank offers apply here?"
  ]
}