Skip to content
Leon Achata
All work
Data

PropFinder AI

A conversational real-estate assistant that turns chat into safe, parameterized SQL.

The problem

Property search forms have dozens of filters. Buyers would rather just say what they want.

How it works

The assistant extracts filters from the conversation — district, area, budget, bedrooms and extras like pets or a balcony — asks for anything missing, then builds a parameterized query against PostgreSQL. The LLM never writes raw SQL, so injection is ruled out by design.

Pipeline

  1. 01Chat
  2. 02Extract filters
  3. 03Validate
  4. 04Build SQL
  5. 05Top 5 matches

Highlights

  • Parameterized queries only — the model never writes SQL
  • Follow-up questions for missing filters
  • Session memory across the conversation
  • Async PostgreSQL with connection pooling

Stack

OpenAIFastAPIPydanticPostgreSQLasyncpg

Want something like this for your company?

I build production versions of these systems on your data and your infrastructure.

Related service: Talk to your data

Book a call