Property search bot

This demo asks a practical question: can a small crew of AI agents research Bangalore suburbs for real-estate investment? The target output includes low, average and high prices, possible rental yield, and useful neighbourhood context. The stack is intentionally simple—CrewAI coordinates the work, LangChain and ChatOpenAI provide the model layer, and a search tool supplies browsable web results.

The important design choice is the split between a researcher and a writer. The researcher receives a role, goal, backstory, tools and a tightly described task. It searches broadly, then narrows into locations such as Indiranagar. The writer receives the collected material and reshapes it into the requested report. That division makes the workflow easier to inspect than one giant prompt because each task has an owner and a visible output.

The run also exposes the limit of this approach. Search results can repeat, miss rental-yield data, or present numbers without enough validation. This is a research pipeline, not a facts pipeline. Its value is in collecting candidate evidence and producing a structured first pass; prices, yields and investment claims still need source checks before anyone acts on them.